Building a precise Frame Cache is a tough job, and is doomed to fail
when attempting to tie cache invalidation to state changes. The only
viable path is to create a system of systematic tagging of processing
steps, and use this as foundation for chained hash values, linked
in accordance to the actual processing structure.
This is complicated by the secondary concern of maintaining memory efficacy
for the render node model, which can be expected to grow to massive scale.
And even while this invocation can not be fully devised right now,
an attempt can be made to build a foundation that is not outright
wasteful, by detaching the logical information from the specific
weaving pattern used for implementation, and by minimising the
representation in memory and computing the compound information
on-demand....
The immediate next goal is to verify properties of render nodes
generated by the builder framework; two kinds of validations
can be distinguished
* structural aspects of the wiring
* the fact that processing functionality is invoked in proper order
Looking into the structural aspects brings about the necessity
to identify the actual processing function bound into some functor.
Some recapitulation of goals and requirements revealed, that this
can not be a merely technical identity record — because the intention
is to base the ''cache key'' on chained processing node identities,
so that the key is stable as long as the user-visible results will be
equivalent. And while structural data can be aggregated, at the
core this information must be provided by the scheme embedded
into the domain ontology, which is tasked with invoking the
builder in order to implement a ''specific processing-asset''
Review the achievements from the last days and map out the further path
for test-driven build-up of a render-node network and invocation.
Notably ''several layers of prototyping'' are in the works now;
it is important to understand the purpose of each such round of
prototyping and to draw the necessary conclusions after closing out.
The next topic to investigate relates to the ''identity'' of nodes and
ports within nodes; this entails to generate a ''symbolic spec'' that
can be verified and used as base for a systematic hash-ID and cache-key...
Since it would in fact be possible to access and write beyond the configured storage,
simply by using the builder API without considering consistency,
it seems advisable to use explicit runtime checks here, instead of
only assertions, and to throw an exception when violating bounds.
Moreover, unsuccessfully attempted to better arrange the functionality
between PortBuilder and WeavingBuilder; seemingly we have an rather tight
coupling here, and also the expectations regarding the processing function
seem to be too tight (but that's the reason why it's an prototype...)
...which then also allow to fill in the missing parts for the
default 1:1 wiring scheme, which connects each »input slot«
of the processing function with the corresponding ''lead node''
- the chaining constructor is picked reliably when the
slicing is done by a direct static_cast
- the function definition can be passed reliably in all cases
after it has been ''decayed,'' which is done here simply by
taking it by-value. This is adequate, since the function
definition must be copied / inlined for each invocation.
With these fixes, the simplest test case now for the first time
**runs through without failure**
This change allows to disentangle the usages of `lib::SeveralBuilder`,
so that at any time during the build process only a single instance is
actively populated, all in one row — and thus the required storage can
either be pre-allocated, or dynamically extended and shrinked (when
filling elements into the last `SeveralBuilder` currently activated)
By packaging into a λ-closure, the building of the actual `Port`
implementation objects (≙ `Turnout` instances) is delayed until the
very end of the build process, and then unloaded into yet another
`lib::Several` in one strike. Temporarily, those building functor
objects are „hidden“ in the current stack frame, as a new `NodeBuilder`
instance is dropped off with an adapted type parameter (embedding the
λ-type produced by the last nested `PortBuilder` invocation, while
inheriting from previous ones.
However, defining a special constructor to cause this »chaining«
poses some challenge (regarding overload resolution). Moreover,
since the actual processing function shall be embedded directly
(as opposed to wrapping it into a `std::function`), further problems
can arise when this function is given as a ''function reference''
...and as expected, this turns up quite some inconsistencies,
especially regarding usage of the »buffer types«.
Basically, the `PortBuilder` is responsible for the high-level functionality
and thus must ensure the nested `WiringBuilder` is addressed and parameterised
properly to connect all »slots« of the processing function.
- can use a helper function in the WiringBuilder to fill in connections
- but the actual buffer types passed over these connectinos are totally
unchecked at that level, and can not see yet how this danger can be
mitigated one level above, where the PortBuilder is used.
- it is still unclear what a »buffer type« actually means; it could
be the pointer type, but it could also imply a class or struct type
to be emplaced into the buffer, which is a special extension to the
`BufferProvider` protocol, yet seems to be used here rather to transport
specific data types required by the actual media handling library (e.g. FFmpeg)
__Analysis__: what kind of verifications are sensible to employ
to cover building, wiring and invocation of render nodes?
Notably, a test should cover requirements and observable functionality,
while ''avoiding direct hard coupling to implementation internals...''
__Draft__: the most simple node builder invocation conceivable...
Code clean-up: mark all buffers with a dedicated tagging type
The point in question is: if we work the LocalTag into the type-hash,
could it be possible to miss an existing entry in the metadata registry?
This could cause two entries to be locked for a single buffer address,
leading to data corruption.
As far as I can see, in the current usage this would not happen,
but unfortunately this problem can not be ruled out, since the BufferProvider
API and protocol is designed to be open for various usage patterns.
However, the same potentially disastrous pattern could also materialise
when registering two different buffer types, and then locking each
for the same buffer location.
...this is a surprisingly tricky issue, since it undercuts the
generic and recursive implementation of buffer handling;
fortunately I've foreseen such demands may arise down the road
and I've reserved an »Local Key« (now renamed into `LocalTag`),
whose meaning is implementation defined and interpreted by
the specific `BufferProvider`
It became clear that a secondary system of connections must be added,
running top-down from a global model context, and thus contrary to the
regular orientation of the node network, which connects upwards from
predecessor to successor, in accordance with the pull principle.
If we accept this wiring as part of the primary structure, it can be
established immediately while building the nodes, thus adding a preconfigured
''pattern of Buffer Descriptors'' to each node, since there is no further
''moving part'' — beyond the wiring to the `BufferProvider`, which thus
becomes part of a global `ModelContext`
As an immediate consequence, the storage for this configuraion should
also be switched to `lib::Several` and handled similar to the primary
node wiring in the Builder...
* conduct analysis regarding allocator handling in the Builder
* turns out we'll have to keep around two different allocators while building
* ⟹ establish the goal to confine usage of the Node allocator to the lower Levels
* consequently must open up the `lib::SeveralBuilder` to be usable
as an intermediary data structure, while building up the target data
* in the initial design, the `SeveralBuilder` was kept opaque, since
contents can be expected to be re-located frequently and thus exposing
elements and taking references could be dangerous — yet this is also
true for `std::vector` however, so people are assumed to know
when they want to shoot themselves into their own foot
...especially what is necessary to represent at this level and what information
is implicit; notably there will be an implicit default wiring, but we allow
for case-by-case deviations
To escape a possible deadlock in analysis, I resort to developing
some kind of free-wheeling presupposition how the **Builder** could
be implemented — a centrepiece of the Lumiera architecture envisioned
thus far — which ''unfortunately'' can only be planned and developed
in a more solid way ''after'' the current »Vertical Slice« is completed.
Thus I find myself in the uncomfortable situation of having to work towards
a core piece, which can not yet be built, since it relies heavily on
the very structures to be built...
...and this line of analysis brings us deep into the ''Buffer Provider''
concept developed in 2012 — which appears to be very well to the point
and stands the test of time.
Adding some ''variadic arguments'' at the right place surprisingly leads
to an ''extension point'' — which in turn directly taps into the
still quite uncharted territory interfacing to a **Domain Ontology**;
the latter is assumed to define how to deal with entities and relationships
defined by some media handling library like e.g. FFmpeg.
So what we're set to do here is actually ''ontology mapping....''
The immediate next step is to build some render nodes directly
in a test setting, without using any kind of ''node factory.''
Getting ahead with this task requires to identify the constituents
to be represented on the first code layer for the reworked code
(here ''first layer'' means any part that are ''not'' supplied
by generic, templated building blocks).
Notably we need to build a descriptor for the `FeedManifold` —
which in turn implies we have to decide on some fundamental aspects
of handling buffers in the render process.
To allow rework of the `ProcNode` connectivity, a lot of presumably obsoleted
draft code from 2011 has to be detached, to be able to keep it in-tree
for further reference (until the rework and refactoring is settled).
As outlined in #1367, the integration effort requires some rework
of existing code, which will be driven ahead by the `NodeLinkage_test`
* redefine Node Connectivity
* build simple `ProcNode` directly in scope
* create an `TurnoutSystem` instance
* perform a ''dummy Node-Invocation''
As a replacement for the `RefArray` a new generic container
has been implemented and tested, in interplay with `AllocationCluster`
* the front-end container `lib::Several<I>` exposes only a reference
to the ''interface type'' `I`, while hiding any storage details
* data can only be populated through the `lib::SeveralBuilder`
* a lot of flexibility is allowed for the actual element data types
* element storage is maintained in a storage extent, managed through
a custom allocator (defaulting to `std::allocator` ⟹ heap storage)
The `SeveralBuilder` employs the same tactic as `std::vector`,
by over-allocating a reserve buffer, which grows in exponential
increments, to amortise better the costs of re-allocation.
This tactic does not play well with space limited allocators
like `AllocationCluster` however; it is thus necessary to provide
an extension point where the actuall allocator's limitation can be
queried, allowing to use what is available as reserve, but not more.
With these adaptations, a full usage cycle backed by `AllocationCluster`
can be demonstrated, including variations of dynamic allocation adjustment.
...identified as part of bug investigation
* make clear that reserve() prepares for an absolute capacity
* clarify that, to the contrary, ensureStorageCapaciy() means the delta
Moreover, it turns out that the assertion regarding storage limits
triggers frequently while writing the test code; so we can conclude
that the `AllocationCluster` interface lures into allocating without
previous check. Consequently, this check now throws a runtime exception.
As an aside, the size limitation should be accessible on the interface,
similar to `std::vector::max_size()`
- decided to allow creating empty lib::Several;
no need to be overly rigid in this point,
since it is move-assignable anyway...
- populate with enough elements to provoke several reallocations
with copying over the existing elements
- precisely calculate and verify the expected allocation size
- verify the use-count due to dedicated allocator instances
being embedded into both the builder and hidden in the deleter
- move-assign data
- all checksums go to zero at end
The setup for `ArrayBucket` is special, insofar it shell de-allocate itself,
which creates the danger of re-entrant calls, or to the contrary, the danger
to invoke this clean-up function without actually invoking the destructor.
These problems become relevant once the destructor function itself is statefull,
as is the case when embedding a non-trivial, instance bound allocator
to be used for the clean-up work. Using the new `lib::TrackingAllocator`
highlighted this potential problem, since the allocator maintains a use-count.
Thus I decided to move the »destruction mechanics« one level down into
a dedicated and well encapsulated base class; invoking ArrayBucket's destructor
thereby becomes the only way to trigger the clean-up, and even ElementFactory::destroy()
can now safely check if the destructor was already invoked, and otherwise
re-invoke itself through this embedded destructor function. Moreover,
as an additional safety measure, the actual destructor function is now
moved into the local stack frame of the object's destructor call, removing
any possibility for the de-allocation to interfere with the destructor
invokation itself
part of the observed deviation stems form bugs in logging and checksum calculation;
but there seems to be a real problem hidden in the allocator usage of the
new component, since the use-cnt of the handle does not drop to zero
While there might be the possibility to use the magic of the standard library,
it seems prudent rather to handle this insidious problem explicitly,
to make clear what is going on here.
To allow for such explicit alignment handling, I have now changed the
scheme of the storage definition; the actual buffer now starts ''behind''
the `ArrayBucket<I>` object, which thereby becomes a metadata managing header.
__To summarise the problem__: since we are maintaining a dynamically sized buffer,
and since we do not want to expose the actual element type through the
front-end object, we're necessarily bound to perform a raw-memory allocation.
This is denoted in bytes, and thus the allocator can no longer manage
the proper alignment automatically. Rather, we get a storage buffer with
just ''some accidental'' alignment, and we must care to request a sufficient
overhead to be able to shift the actual storage area forward to the next
proper alignment boundary. Obviously this also implies that we must
store this individual padding adjustment somewhere in the metadata,
in order to be able to report the correct size of the block later
on de-allocation.
The solution implemented thus far turns out to be not sufficient
for ''over-aligned-data'', as the raw-allocator can not perform the
''magic work'' because we're exposing only `std::byte` data.
This adaptor works in concert with the generic allocator
building blocks (prospective ''Concepts'') and automatically
registers a either static or dynamic back-link to the factory
for clean-up.
Use this wrapper fore more in-depth test of the new `TrackingAllocator`
and verify proper behaviour through the `EventLog`
- ability to verify a hash-checksum
- ability to watch number of allocations and allotted bytes
- using either a common global pool or a separate dedicated pool
- log all operations into a common `EventLog` instance
- front-end adaptors for use as C++ custom allocator
...these features are now used quite regularly,
and so a dedicated documentation test seems indicated.
Actually my intention is to add a tracking allocator to these test helpers
(and then to use that to verify the custom allocator usage of `lib::Several`)
Phew... this was a tough one — and not sure yet if this even remotely works...
Anyway, the `lib::SeveralBuilder` is already prepared for collaboration with a
custom allocator, since it delegates all memory handling through a base policy,
which in turn relies on std::allocator_traits.
The challenge however is to find a way...
* to make this clear and easy to use
* to expose an extension point for specific tweaks
* and to make all this work without excessive header cross dependencies
This is a low-level interface to allow changing the size of
the currently latest allocation in `AllocationCluster`; a client
aware of this capability can perform a real »in-place re-alloc«,
assuming the very specific usage constraints can be met.
`lib::Several<X>` will use this feature when attached to an
`AllocationCluster`; with this special setup, an previously
unknown number of non-copyable objects can be built without
wasting any storage, as long as the storage reserve in the
current extent of the `AllocationCluster` is sufficient.
...use some pointer arithmetic for this test to verify
some important cases of object placement empirically.
Note: there is possibly a very special problematic case
when ''over aligned objects'' are not placed in accordance
to their alignment requirements. Fixing this problem would
be non-trivial, and thus I have only left a note in #1204
...including the interesting cases where objects are relocated
and the element spread is changed. With the help of the checksum
feature built into the test-dummy objects, the properly balanced
invocation of constructors can be demonstrated
PS: for historical context...
Last week the "Big F**cking Rocket" successfully performed the
test flight 4; both booster and Starship made it back to the
water surface and performed a soft splash-down after decelerating
to speed zero. The Starship was even able to maintain control
in spite of quite some heat damage on the steering flaps.
Yes ... all techies around the world are thrilled...
- spread change now retains the nominal element reserve
- `capacity()` and `capReserve()` now exposed on the builder API
- factor out the handling check safety functions
- rewrite the `resize()` builder function to be more generic
__Test now covers__ example with trivial data type, which can
indeed be resized and allows to grow buffer on-the fly without
requiring any knowledge of the actual type (due to using `memmove`)
building on the preceding analysis, we can now demonstrate that
the container is initially able to grow, but looses this capability
after accepting one element of unknown subclass type...
`lib::Several` is designed to be highly adaptable, allowing for
several quite distinct usage styles. On the downside, this requires
to perform some checks at runtime only, since the ability to handle
some element depends on specific circumstances.
This is a notable difference to `std::vector`, which is simply not capable
of handling ''non-copyable'' types, even if given an up-front memory reservation.
The last test case provided with the previous changeset did not trigger
an exception, but closer investigation revealed that this is correct,
since in this specific situation the container can accept this object type,
thereby just loosing the ability to move-relocate further objects.
A slightly re-arranged test scenario can be used to demonstrate this fine point.
- the test-dummy objects need a `noexcept` move ctor
- **bug** here: need an explicit check to prevent other types
than the known element type from ''sneaking in''
The `SeveralBuilder` is very flexible with respect to added elements,
but it will investigate the provided type information and reject any
further build operation that can not be carried out safely.
...turns out that we must ensure to pass a plain "object" type
to the standard allocator framework (no const, no references).
Here, ''object in C++ terminology'' means a scalar or record type,
but no functor, no references and no void,
Consider what (not) to support.
Notably I decided ''not to support'' moving out of an iterator,
since doing so would contradict the fundamental assumptions of
the »Lumiera Forward Iterator« Concept.
Start verifying some variations of element placement,
still focussing on the simple cases
Parts of the decision logic for element handling was packaged
as separate »strategy« class — but this turned out to be neither
a real abstraction, nor configurable in any way. Thus it is better
to simplify the structure and turn these type predicates into simple
private member functions of the SeveralBuilder itself
...and the nice thing is, the recently built `IterIndex` iteration wrapper
covers this functionality right away, simply because `lib::Several`
is a generic container with subscript operator.
...passes the simplest unit test
* create a Several<int>
* populate from `std::initializer_list`
* random-access to elements
''next step would be to implement iteration''
Some decisions
- use a single template with policy base
- population via separate builder class
- implemented similar to vector (start/end)
- but able to hold larger (subclass) objects
- basically works out-of-the-box now
- the hard wired fixed Extent size is a serious limitation
- however, this is not the intended primary use, rather complementary
...this is an important detail: quite commonly, a custom allocator
is actually implemented as monostate, to avoid bloating every client container
with a backlink pointer; by inheriting the `StdFactory` adapter from the
allocator, the empty-base optimisation can be exploited.
In the standard case thus LinkedElements is the same size as a single
pointer, which is already exploited at several places in the code base.
Notably `AllocationCluster` uses a »virtual overlay« to dress-up the
position pointer as `LinkedElements`, allowing to delegate most of the
administration and memory management to existing and verified code.
With this adjustments, `LinkedElements` pass the tests again
and the rework of `AllocationCluster` is considered complete.
This is the first validation of the new design:
the policy to take ownership can be reimplemented simply
by delegating to the adaptor for a C++ standard allocator
...what I've implemented yesterday is effectively the same functionality
as provided automatically by the C++ object system when using a virtual destructor.
Thus a much cleaner solution is to turn `Destructor` into a interface
and let C++ do all the hard work.
Verified in test: works as intended
This is the first draft, implementing the invocation explicitly
through a trampoline function. While it seems to work,
the formulation can probably be simplified....
- rather accept hard-wired limits than making the implementation excessively generic
- by exploiting the layout, the administrative overhead can be reduced significantly
- the trick with the "virtual managment overlay" allows to hand-off most of the
clean-up work to C++ destructor invocation
- it is important to verify these low-level arrangements explicitly by unit-test
...due to the decision to use a much simpler allocation scheme
to increase probability for actual savings, after switching the API
and removing all trading related aspects, a lot of further code is obsoleted
Notably this raises the difficult question,
whether to ensure **invocation of destructors**.
Not invoking dtors ''breaks one of the most fundamental contracts''
of the C++ language — yet the infrastructure to invoke dtors in such
a heterogeneous cluster of allocations creates a hugely significant
overhead and is bound to poison the caches (objects to be deallocated
typically sit in cold memory pages).
What makes this decision especially daunting is the fact that the
low-level-Model can be expected to be one of the largest systemic
data structures (letting aside the media buffers).
I am leaning towards a compromise: turn down this decision
towards the user of the `AllocationCluster`
At the time of the initial design attempts, I naively created a
classic interface to describe an fixed container allocated ''elsewhere.''
Meanwhile the C++ language has evolved and this whole idea looks
much more as if it could be a ''Concept'' (C++20). Moreover, having
several implementations of such a container interface is deemed inadequate,
since it would necessitate ''at least two indirections'' — while
going the Concept + Template route would allow to work without any
indirection, given our current understanding that the `ProcNode` itself
is ''not an interface'' — rather a building block.
- the starting point is the idea to build a dedicated ''turnout system''
- `StateAdapter`, `BuffTable` ⟶ `FeedManifold` and _Invocation_ will be fused
- actually, the `TurnoutSystem` will be ''pulled'' and orchestrate the invocation
- the structure is assumed to be recursive
The essence of the Node-Invocation, as developed 2009 / 2011 remains intact,
yet it will be organised along a clearer structure
Facing quite some difficulties here, since there are (at least)
two abandoned past efforts towards building a render node network
in the code base; the structure and architecture decisions from these
previous attempts seem largely valid still, yet on a technical level,
the style of construction evolved considerably in the meantime. Moreover,
these old fragments of code, written during the early stages of the
project, were lacking clear goals and anchor points at places;
the situation is quite different now in this respect.
Sticking to well proven practice, the rework will be driven by a test setup,
and will progress over three steps with increasing levels of integration.
The initial effort of building a Scheduler can now be **considered complete**
Reaching this milestone required considerable time and effort, including
an extended series of tests to weld out obvious design and implementation flaws.
While the assessment of the new Scheduler's limitation and traits is ''far from complete,''
some basic achievements could be confirmed through this extended testing effort:
* the Scheduler is able to follow a given schedule effectively,
until close up to the load limit
* the ''stochastic load management'' causes some latency on isolated events,
in the order of magnitude < 5ms
* the Scheduler is susceptible to degradation through Contention
* as mitigation, the Scheduler prefers to reduce capacity in such a situation
* operating the Scheduler effectively thus requires a minimum job size of 2ms
* the ability for sustained operation under full nominal load has been confirmed
by performing **test sequences with over 80 seconds**
* beyond the mentioned latency (<5ms) and a typical turnaround of 100µs per job
(for debug builds), **no further significant overhead** was found.
Design, Implementation and Testing were documented extensively in the [https://lumiera.org/wiki/renderengine.html#Scheduler%20SchedulerProcessing%20SchedulerTest%20SchedulerWorker%20SchedulerMemory%20RenderActivity%20JobPlanningPipeline%20PlayProcess%20Rendering »TiddlyWiki« #Scheduler]
This test completes the stress-testing effort
and summarises the findings
* Scheduler performs within relevant parameter range without significant overhead
* Scheduler can operate with full load in stable state, with 100% correct result
The behaviour seems consistent and the schedule breaks at the expected point.
At first sight, concurrency seems slightly to low; detailed investigation
however shows that this is due to the structure of the load graph,
and in fact the run time comes close to optimal values.
the `BreakingPoint` tool conducts a binary search to find the ''stress factor''
where a given schedule breaks. There are some known deviations related to the
measurement setup, which unfortunately impact the interpretation of the
''stress factor'' scale. Earlier, an attempt was made, to watch those factors
empirically and work a ''form factor'' into the ''effective stress factor''
used to guide this measurement method.
Closer investigation with extended and elastic load patters now revealed
a strong tendency of the Scheduler to scale down the work resources when not
fully loaded. This may be mistaken by the above mentioned adjustments as a sign
of a structural limiation of the possible concurrency.
Thus, as a mitigation, those adjustments are now only performed at the
beginning of the measurement series, and also only when the stress factor
is high (implying that the scheduler is actually overloaded and thus has
no incentive for scaling down).
These observations indicate that the »Breaking Point« search must be taken
with a grain of salt: Especially when the test load does ''not'' contain
a high degree of inter dependencies, it will be ''stretched elastically''
rather than outright broken. And under such circumstances, this measurement
actually gauges the Scheduler's ability to comply to an established
load and computation goal.
...this seems to be the last topic for this investigation of Scheduler behaviour;
the goal is to demonstrate readiness for stable-state operation over an extended period of time
- use parameters known to produce a clean linear model
- assert on properties of this linear model
Add extended documentation into the !TiddlyWiki,
with a textual account of the various findings,
also including some of the images and diagrams,
rendered as SVG
This amends test code, which was commented-out for some time,
and was affected by the changes in load-graph generation:
a983a506b
These changes typically lead to a simplified topology at the end
of the load graph, since open ends are no longer connected to a
single exit node. In the case here, level 27 is no longer generate,
and level 26 is now comprised of three nodes, two of them with load=2
Investigate the behaviour over a wider range of job loads,
job count and worker pool sizes. Seemingly the processing
can not fully utilise the available worker pool capacity.
By inspection of trace-dumps, one impeding mechanism could
be identified: the »stickiness« of the contention mitigation.
Whenever a worker encounters repeated contention, it steps up
and adds more and more wait cycles to remove pressure from the
schedule coordination. As such this is fine and prevents further
degradation of performance by repeated atomic synchronisation.
However, this throttling was kept up needlessly after further
successful work-pulls. Since job times of several milliseconds
can be expected on average in media processing, such a long
retention would spread a performance degradation over a duration
of several frames. Thus, the scheme for step-down was changed
to decrease the throttling by a power series rather than just
documenting the level.
Use the statistic functions imported recently from Yoshimi-test
to compute a linear regression model as immediate test result.
Combining several measurement series, this allows to draw conclusions
about some generic traits and limitations of the scheduler.
Visual tweaks specific to this measurement setup
* include a numeric representation of the regression line
* include descriptive axis labels
* improve the key names to clarify their meaning
* heuristic code for the x-ticks
Package these customisations as a helper function into the measurement tool
After a lot of further tinkering, seemingly arriving at a
somewhat satisfactory solution for the layout and arrangement of
test definitions and especially the table for measurement series.
While the complete setup remains fragile indeed, and complexity is more
hidden than reduced — the pragmatic compromise established yesterday
at least allows to reduce the amount of boilerplate in the test or
measurement setup to make the actual specifics stand out clearly.
----
As an aside, the usage of the `DataFile` type imported from Yoshimi-test
recently was re-shaped more towards a generic handling of tabular data with
CSV storage option; thus renaming the type now into `DataTable`.
Persistent storage is now just one option, while another usage pattern
compounds observation data into table rows, which are then directly
rendered into a CSV string, e.g. for visualisation as Gnuplot graph.
Encountering ''just some design problems related to the test setup,''
which however turn out hard to overcome. Seems that, in my eagerness
to create a succinct and clear presentation of the test, I went into
danger territory, overstretching the abilities of the C++ language.
After working with a set of tools created step by step over an extended span of time,
''for me'' the machinations of this setup seem to be reduced to flipping a toggle
here and there, and I want to focus these active parts while laying out this test.
''This would require'' to create a system of nested scopes, while getting more and more
specific gradually, and moving to the individual case at question; notably any
clarification and definition within those inner focused contexts would have to be
picked up and linked in dynamically.
Yet the C++ language only allows to be ''either'' open and flexible towards
the actual types, or ''alternatively'' to select dynamically within a fixed
set of (virtual) methods, which then must be determined from the beginning.
It is not possible to tweak and adjust base definitions after the fact,
and it is not possible to fill in constant definitions dynamically
with late binding to some specific implementation type provided only
at current scope.
Seems that I am running against that brick wall over and over again,
piling up complexities driven by an desire for succinctness and clarity.
Now attempting to resolve this quite frustrating situation...
- fix the actual type of the TestChainLoad by a typedef in test context
- avoid the definitions (and thus the danger of shadowing)
and use one `testSetup()` method to place all local adjustments.
With the addition of a second tool `bench::ParameterRange`,
the setup of the test-context for measurement became confusing,
since the original scheme was mostly oriented towards the
''breaking point search.''
On close investigation, I discovered several redundancies, and
moreover, it seems questionable to generate an ''adapted-schedule''
for the Parameter-Range measurement method, which aims at overloading
the scheduler and watch the time to resolve such a load peak.
The solution entertained here is to move most of the schedule-ctx setup
into the base implementation, which is typically just inherited by the
actual testcase setup. This allows to leave the decision whether to build
an adapted schedule to the actual tool. So `bench::BreakingPoint` can
always setup the adapted schedule with a specific stress-factor,
while `bench::ParameterRange` by default does nothing in this
respect, and thus the `ScheduleCtx` will provide a default schedule
with the configured level-duration (and the default for this is
lowered to 200µs here).
In a similar vein, calculation of result data points from the raw measurement
is moved over into the actual test setup, thereby gaining flexibility.
Rework the existing tool to capture the measurement series
into the newly integrated CSV-based data storage, allowing
to turn the results into a Gnuplot-visualisation.
...which is added automatically whenever additional data columns are present
Result can only be verified visually
* the upper diagram should show the first fibonacci points
* a (correct) linear regression line should be overlayed in red
* below, a secondary diagram should appear, with aligned axis
* the row "one" in this diagram should be shown as impulses
* the further rows "two" and "three" should be drawn as
green points, using the secondary Y-axis (values 100-250)
* Gnuplot can handle missing data points
The idea is to build the Layout-branching into the generated Gnuplot script,
based on the number of data columns detected. If there is at least one further
data column, then the "mulitplot" layout will be used to feature this
additional data in a secondary diagram below with aligned axis;
if more than one additional data column is present, all further
visualisation will draw points, using the secondary Y-axis
Moreover, Gnuplot can calculate the linear regresssion line itself,
and the drawing will then be done using an `arrow` command,
defining a function regLine(x) based on the linear model.
- `forElse` belongs to the metaprogramming utils
- have a CSVLine, which is a string with custom appending mechanism
- this in turn allows CSVData to accept arbitrary sized tuples,
by rendering them into CSVLine
the metafunction `is_basically<X>` performed only an equality match,
while, given it's current usage, it should also include a subtype-interface-match.
This changes especially the `is_StringLike<S>` metafunction,
both on const references and on classes built on top of string
or string_view.
Whenever a class defines a single-arg templated constructor,
there is danger to shadow the auto-generated copy operations,
leading to insidious failures.
Some months ago, I did the ''obvious'' and added a tiny helper,
allowing to mask out the dangerous case when the ''single argument''
is actually the class itself (meaning, it is a copy invocation and
not meant to go through this templated ctor...
As this already turned out as tremendously helpful, I now extended
this helper to also cover cases where the problematic constructor
accepts variadic arguments, which is quite common with builder-helpers
The intention is to create a library of convenient building blocks;
providing a visualisation should be as simple as invoking a free function
with CSV data, yet with the ability to tweak some lables or display
variations if desired.
This can be achieved by..
* having a series of ready-made standard visualisations
* expose a function call for each, accepting a data-context builder
* provide secondary convenience shortcuts, which add some of the expected bindings
* notably a shortcut is provided to take the data as CSV-string
* augmented by a wrapper/builder to allow defining data points inline
Deliberately keep it unstructured and add dedicated functions
for each new emerging use case; hopefully some commen usage scheme
will emerge over time.
* Data is to be handed in as an iterator over CSV-strings.
* will have to find out about additional parametrisation on a case-by-case base
A minimalist `TextTemplate` engine is available for in-project use.
* supports only the bare minimum of features (no programming language)
* substitution of `${placeholder}` by key-name data access
* conditional section `${if key}...${end if}`
* iteration over a data sequence
* other then most solutions available as library,
this implementation does **not require** a specific data type,
nor does it invent a dynamic object system or JSON backend;
rather, a generic ''Data Source Adapter'' is used, which can
be specialised to access any kind of ''structured data''
* the following `DataSource` specialisations are provided
* `std::map<string,string>`
* Lumiera »External Tree Description« (based on `GenNode`)
* a string-based spec for testing
This extension is required to use GenNode as data source for text-template instantiation.
I am aware that such a function could counter the design intent for GenNode,
because it could be (ab)used to "just get the damn value" and then
parse back the results...
...turns out challenging, since our intention here
is borderline to the intended design of the Lumiera ETD.
It ''should work'' though, when combined with a Variant-visitor...
Document existing data binding logic and investigate in detail
what must be done to enable a similar binding backed by Lumiera's ETD structures.
This analysis highlights some tricky aspects, which can be accommodated by
slight adjustments and generalisations in the `TextTemplate` implementation
* `GenNode` is not structured string data, rather binary data
* thus exposing a std::string_view is not adequate, requiring to
pick up the result type from the actual data binding
* moreover, to allow for arbitrary nested scopes, a back-pointer
to the parent scope must be maintained, which requires stable memory locations.
This can best be solved within the InstanceCore itself, which manages
the actual hierarchy of data source references.
* the existing code happens already to fulfil this requirement, but
for sake of clarity, handling of such a nested scope is now extracted
into a dedicated operation, to highlight the guaranteed memory layout.
...hoped to keep it simple, but this is inevitable, since we
want to provide a CSV list as value within a list of key=value
bindings, and all packaged into a simple string for easy testing.
Thus the parsing RegExp just needs two branches for simple and quoted vals
...implemented by simply parsing the string into key=value pairs,
which are then stored into a shared map. The actual data binding
implementation can thus be inherited from the existing Map-binding
While they were detected just fine, thy were passed-through
unaltered, which subverts the purpose of such an escape,
which is to allow for the tag syntax to be present in the
processed, substituted document (e.g. when generating a
shell script)
thus `\${escaped}` becomes `${escaped}`
...turns out the ''pipeline design'' is not a good fit for the
Action compilation, since the compiler needs to refer to previous Actions;
better to let the compiler ''build'' the `ActionSeq`
...implemented as »custom processing layer« within a
demand-driven parsing pipeline, with the ability to
inject additional Action-tokens to represent the intermittent
constant text between tags; special handling to expose one
constant postfix after the last active tag.
MatchSeq was imported recently from the Yoshimi-testsuite,
as supporting helper for the CSV table component.
Actually this is just a thin wrapper on top of std::regex_iterator,
which in turn has properties and behaviour very similar to Lumiera's
»Forward Iterator« concept (in fact, it was a source of inspiration to
generalise such a pattern).
So this is an obvious round out and cleanup, as it requires just some
minor additions and adjustments to allow processing a sequence of matches
through a for-loop or some elaborate pipelining setup.
The way I've written this helper template, as a byproduct
it is also possible to maintain the back-refrence to the container
through a smart-ptr. In this case, the iterator-handle also manages
the ownership automatically.
...mostly we want the usual convenient handling pattern for iterators,
but with the proviso actually to perform an access by subscript,
and the ability to re-set to another current index
* establish the feature set to provide
* choose scheme for runtime representation
* break down analysis to individual parsing and execution steps
* conclude which actions to conduct and the necessary data
* derive the abstract binding API required
Conducted an extended investigation regarding text templating
and the library solutions available and still maintained today.
The conclusion is
* there are some mature and widely used solutions available for C++
* all of these are considered a mismatch for the task at hand,
which is to generate Gnuplot scripts for test data visualisation
Points of contention
* all solutions offer a massive feature set, oriented towards web content generation
* all solutions provide their own structured data type or custom property-tree framework
**Decision** 🠲 better to write a minimalistic templating engine from scratch rather
In the Lumiera code base, we use C-String constants as unique error-IDs.
Basically this allows to create new unique error IDs anywhere in the code.
However, definition of such IDs in arbitrary namespaces tends to create
slight confusion and ambiguities, while maintaining the proper use statements
requires some manual work.
Thus I introduce a new **standard scheme**
* Error-IDs for widespread use shall be defined _exclusively_ into `namespace lumiera::error`
* The shorthand-Macro `LERR_()` can now be used to simplify inclusion and referral
* (for local or single-usage errors, a local or even hidden definition is OK)
reduce footprint of lib/util.hpp
(Note: it is not possible to forward-declare std::string here)
define the shorthand "cStr()" in lib/symbol.hpp
reorder relevant includes to ensure std::hash is "hijacked" first
showDecimal -> decimal10 (maximal precision to survive round-trip through decimal representation=
showComplete -> max_decimal10 (enough decimal places to capture each possible distinct floating-point value)
Use these new functions to rewrite the format4csv() helper
...this uncovered one inconsistency: when directly adding values
into one of the embedded data vectors, the inconsistent size
was allowed to persist even when adding / removing lines.
This is in contradiction to the behavior for the CSV dump,
which uses index positions from the front of all vectors uniformely.
Thus changed the behaviour of adding a new row, so that it now
caps all vectors to a common size
also added function to clear the table
verify also that clean-up happens in case of exceptions thrown;
as an aside, add Macro to check for ''any'' exception and match
on something in the message (as opposed to just a Lumiera Exception)
...using the same method for sake of uniformity
Also move the permissions helpers to the file.hpp support functions
and setup a separate unit test for these
Inspired by https://stackoverflow.com/a/58454949
Verified behaviour of fs::create_directory
--> it returns true only if it ''indeed could create'' a new directory
--> it returns false if the directory exists already
--> it throws when some other obstacle shows up
As an aside: the Header include/limits.h could be cleaned up,
and it is used solely from C++ code, thus could be typed, namespaced etc.
Since this is a much more complicated topic,
for now I decided to establish two instances through global variables:
* a sequence seeded with a fixed starting value
* another sequence seeded from a true entropy source
What we actually need however is some kind of execution framework
to define points of random-seeding and to capture seed values for
reproducible tests.
Relying on random numbers for verification and measurements is known to be problematic.
At some point we are bound to control the seed values -- and in the actual
application usage we want to record sequence seeding in the event log.
Some initial thoughts regarding this intricate topic.
* a low-ceremony drop-in replacement for rand() is required
* we want the ability to pick-up and control each and every usage eventually
* however, some usages explicitly require true randomness
* the ability to use separate streams of random-number generation is desirable
Yesterday I decided to include some facilities I have written in 2022
for the Yoshimi-Testsuite. The intention is to use these as-is, and just
to adapt them stylistically to the Lumiera code base.
However — at least some basic documentation in the form of
very basic unit-tests can be considered »acceptance criteria«
Initially the model was that of a single graph starting
with one seed node and joining all chains into a single exit node.
This however is not well suited to simulate realistic calculations,
and thus the ability for injecting additional seeds and to randomly
sever some chains was added -- which overthrows the assumption of
a single exit node at the end, where the final hash can be retrieved.
The topology generation used to pick up all open ends, in order to
join them explicitly into a reserved last node; in the light of the
above changes, this seems like an superfluous complexity, and adds
a lot of redundant checks to the code, since the main body of the
algorithm, in its current form, already does all the necessary
bound checks. It suffices thus to just terminate the processing
when the complete node space is visited and wired.
Unfortunately this requires to fix basically all node hashes
and a lot of the statistics values of the test; yet overall
the generated graphs are much more logical; so this change
is deemed worth the effort.
Allow easily to generate a Chain-Load with all nodes unconnected,
yet each node on a separate level.
Fix a deficiency in the graph generation, which caused spurious
connections to be added at the last node, since the prune rule
was not checked
...the previous setup produced a single linear chain
instead of a set of unconnected nodes.
With this, the behaviour is more like expected,
but concurrency is still too low
- better use a Test-Chain-Load without any dependencies
- schedule all at once
- employ instrumentation
- use the inner »overall time« as dependent result variable
The timing results now show an almost perfect linear dependency.
Also the inner overall time seems to omit the setup and tear-down time.
But other observed values (notably the avgConcurrency) do not line up
- fill the range randomly with probe points
- use the node count as independent parameter
- measurement method *works as intended*
- results indeed show a linear relationship
Results are ''interesting'' however, since the (par,time) points
seem to be arranged into two lines, implying that about half
of the runs were somehow ''degraded'' and performed way slower.
With the latest improvements, the »breaking point search« works as expected
and yields meaningful data; however — it seems to be well suited rather
for specific setups, which involve an extended graph with massive dependencies,
because only such a setup produces a clearly defined ''breaking point.''
Thus I'm considering to complement this research by another measurement setup
to establish a linear regression model of the Scheduler expense.
To allow integration of this different setup into the existing stress-test-rig,
some rearrangements of the builder notation are necessary; especially we need
to pass the type name of the actual tool, and it seems indicated to
reorder the source code to provide the config base class `StressRig`
at the top, followed by a long (and very technical) implementation
namespace.
It turns out to be not correct using all the divergence in concurrency
as a form factor, since it is quite common that not all cores can be active
at every level, given the structural constraints as dictated by the load graph.
On the other hand, if the empirical work (non wait-time) concurrency
systematically differs from the simple model used for establishing the schedule,
then this should indeed be considered a form factor and deduced from
the effective stress factor, since it is not a reserve available for speed-up
The solution entertained here is to derive an effective compounded sum
of weights from the calculation used to build the schedule. This compounded
weight sum is typically lower than the plain sum of all node weights, which
is precisely due to the theoretical amount of expense reduction assumed
in the schedule generation. So this gives us a handle at the theoretically
expected expense and through the plain weight sum, we may draw conclusion
about the effective concurrency expected in this schedule.
Taking only this part as base for the empirical deviations yields search results
very close to stressFactor ~1 -- implying that the test setup now
observes what was intended to observe...
In binary search, in order to establish the invariant initially,
a loop is necessary, since a single step might not be sufficient.
Moreover, the ongoing adjustments jeopardise detection of the
statistical breaking point condition, by causing a negative delta
due to gradually approaching the point of convergence -- leading
to an ongoing search in a region beyond the actual breaking point.
Various misconceptions identified in the feedback path of the test algorithm.
- statistics are cumulative, which must be incorporated by norming on time base
- average concurrency includes idle times, which is besides the point within this
test setup, since additional wait-phases are injected when reducing stress
Relying on the new instrumentation facility, the actually effective
concurrency and cumulative run time of the test jobs can be established.
These can now be cast into a form-factor to represent actual excess expenses
in relation to the theoretical model.
By allowing to adjust the adapted schedule by this form factor,
it can be made to reflect more closely the actual empiric load,
hopefully leading to a more realistic effect of the stress-factor
and thus results better suited to conclude on generic behaviour.
...turns out rather challenging to come up with any test case,
that is both meaningful, simple to setup and understand, yet still
produces somewhat stable values. `IncidenceCount` seems most valuable
for investigation and direct inspection of results
Various experiments to watch Scheduler behaviour under extended load.
Notably the example committed here makes the Scheduler run for 1.2 sec
and process 800 jobs with 10ms each, thereby putting the system into
100% load on all CPUs
- supplement the pre-dimensioning for data capture; without that,
sporadic memory corruption indeed happens (as expected, since
concurrent re-allocation of the vector with an entry for each
thread is not threadsafe, and this test shows much contention)
- add a top-level logging for better diagnostics of errors
emanating from the test run
Basically users are free to place the measurement calls to their liking.
This implies that bracketed measurement intervals can be defined overlapping
even within a single thread, thereby accounting the overlapping time interval
several times. However, for the time spent per thread, only actual thread
activity should be counted, disregarding overlaps. Thus introduce a
new aggregate, ''active time'', which is the sum of all thread times.
As an aside, do not need explicit randomness for the simple two-thread
test case — timings are random anyway...
+ bugfix for out-of-bounds access
...since we've established already an integration over the event timeline,
it is just one simple further step to determine the concurrency level
on each individual segment of the timeline. Based on this attribution
- the averaged concurrenty within the observation range can be computed as weighted mean
- moreover we can account for the precise cumulated time spent at each concurrency level
...using a simplistic allocation of next-slot based on initialisation
of a thread_local storage. This implies that this helper can not be
reset or reused, and that there can not be multiple or long-lived instances.
Keep-it-simple for now...
...to sort out the interpretation of measurement results,
the actual duration and concurrency of ComputationLoad invocations
should be recorded, allowing to draw conclusions regarding the
Scheduler's performance as opposed to further system and thread
management effects due to concurrent operation under pressure.
After an extended break due to "real life issues"....
Pick up the investigation, with the goal to ascertain a valid definition
and understanding of all test parameters. A first step is to establish
a baseline ''without using a computational load''; this might be some kind
of base overhead of the scheduler.
However -- the way the test scaffolding was built, it is difficult to
create a feedback loop for the statistical test setup with binary search,
since it is not really clear how the single control parameter of the test algorithm,
the so called "stress factor", shall be interpreted and how it can be
combined with a base load.
An extended series of tests, while watching the observed value patterns qualitatively,
seems to corroborate the former results, indicating that the base expense
in my test setup (using a debug build) is at ~200µs / Node / core.
Yet the difficulty to interpret this result and arrive at a logical and generic model
prevents me from translating this into a measurement scheme, which can
be executed independently from a specific test setup and hardware
The goal is to devise a load more akin to the expected real-world processing patterns,
and then to increase the density to establish a breaking point.
Preliminary investigations focus on establishing the properties of this load
and to ensure the actual computation load behaves as expected.
Using the third Graph pattern prepared thus far, which produces
short chains of length=2, yet immediately spread out to maximum concurrency.
This leads to 5.8 Nodes / Level on average.
...as it turned out, this segfault was caused by flaws in the ScheduleCtx
used for generate the test-schedule; especially when all node-spreads are set
to zero and thus all jobs are scheduled immediately at t=0, there was a loophole
in the logic to set the dependencies for the final »wake-up« job.
When running such a schedule in the Stress-Test-Bench, the next measurement run
could be started due to a premature wake-up job, thereby overrunning the previous
test-run, which could be still in the middle of computations.
So this was not a bug in the Scheduler itself, yet something to take care of
later when programming the actual Job-Planning and schedule generation.
Search processing pattern for massive parallel test.
The goal is to get all cores into active processing most of the time,
thus we need a graph with low dependency management overhead, which is
also consistently wide horizontally to have several jobs in working state
all of the time. The investigation aims at finding out about systematic
overheads in such a setup.
This is just another (obvious) degree of freedom, which could be
interesting to explore in stress testing, while probably not of much
relevance in practice (if a job is expected to become runable earlier,
in can as well be just scheduled earlier).
Some experimentation shows that the timing measurements exhibit more
fluctuations, but also slightly better times when pressure is low, which
is pretty much what I'd expect. When raising pressure, the average
times converge towards the same time range as observed with time bound
propagation.
Note that enabling this variation requires to wire a boolean switch
over various layers of abstraction; arguably this is an unnecessary
complexity and could be retracted once the »experimentation phase«
is over.
This completes the preparation of a Scheduler Stress-Test setup.
The `volatile` was used asymmetrically and there was concern that
this usage makes the `ComutationalLoad` dependent on concurrency.
However, an impact could not be confirmed empirically.
Moreover, to simplify this kind of tests, make the `schedDepends`
directly configurable in the Stress-Test-Rig.
...watching those dumps on the example Graph with excessive dependencies
made blatantly clear that we're dispatching a lot of unnecessary jobs,
since the actual continuation is /always/ triggered by the dependency-NOTIFY.
Before the rework of NOTIFY-Handling, this was rather obscured, but now,
since the NOTIFY trigger itself is also dispatched by the Scheduler,
it ''must be this job'' which actually continues the calculation, since
the main job ''can not pass the gate'' before the dependency notification
arrives.
Thus I've now added a variation to the test setup where all these duplicate
jobs are simply omitted. And, as expected, the computation runs faster
and with less signs of contention. Together with the other additional
parameter (the base expense) we might now actually be able to narrow down
on the observation of a ''expense socket'', which can then be
attributed to something like an ''inherent scheduler overhead''
...actually difficult to integrate into the existing scheme,
which is entirely level-based. Can only be added to the individual Jobs,
not to the planning and completion-jobs — which actually shouldn't be a problem,
since it is beneficial to dispatch the planning runs earlier
The next goal is to determine basic performance characteristics
of the Scheduler implementation written thus far;
to help with these investigations some added flexibility seems expedient
- the ability to define a per-job base expense
- added flexibility regarding the scheduling of dependencies
This changeset introduces configuration options
While the idea with capturing observation values is nice,
it definitively does not belong into a library impl of the
search algorithm, because this is usage specific and grossly
complicates the invocation.
Rather, observation data can be captured by side-effect
from the probe-λ holding the actual measurement run.
- Result found in typically 6-7 steps;
- running 20 instead of 30 samples seems sufficient
Breaking point in this example at stress-Factor 0.47 with run-time 39ms
This statistical criterion defines when to count observed Scheduler performance
as loosing control. The test is comprised of three observations, which
all must be confirmed:
- an individual run counts as accidentally failed when the execution slips
away by more than 2ms with respect to the defined overall schedule.
When more than 55% of all observed runs are considered as failed,
the first condition is met
- moreover, the observed standard derivation must also surpass the
same limit of > 2ms, which indicates that the Scheduling mechanism
is under substantial strain (on average, the slip is ~ 200µs)
- the third condition is that the ''averaged delta'' has surpassed
4ms, which is 2 times the basic failure indicator.
These conditions are based on watching the Scheduler in operation;
typically all three conditions slip away by large margin after a
very narrow yet critical increase in the stress level.
Using three conditions together should improve robustness; often
the problems to keep up with the schedule build up over some parameter
range, yet the actual decision should be based on complete loss of control.
adapt the code written yesterday explicitly for the test case
into the new framework for performing a stress-test run.
Notable difference: times converted to millisecond immediately
Elaborate the draft to include all the elements used directly in the test case thus far;
the goal is to introduce some structuring and leave room for flexible confguration,
while implementing the actual binary search as library function over Lambdas.
My expectation is to write a series of individual test instances with varying parameters;
while it seems possible to add further performance test variations into that scheme later on.
- the goal is to run a binary search
- the search condition should be factored out
- thus some kind of framework or DSL is required,
to separate the technicalities of the measurement
from the specifics of the actual test case.
- repeated invocations of the same test setup for statistics
- the usual nasty 64-node graph with massive fork out
- limit concurrency to 4 cores
- tabulate data to look for clues regarding a trigger criteria
Hypothesis: The Scheduler slips off schedule when all of the
following three criteria are met:
- more than 55% glitches with Δ > 2ms
- σ > 2ms
- ∅Δ > 4ms
...this one was quite silly: obviously we need a separate instance
of the memory block ''per invocation'', otherwise concurrent invocations
would corrupt each other's allocation. The whole point of this variant
of the computation-load is to access a ''private'' memory block...
- schedule can now be adapted to concurrency and expected distribution of runtimes
- additional stress factor to press the schedule (1.0 is nominal speed)
- observed run-time now without Scheduler start-up and pre-roll
- document and verify computed numbers
...based on the adapted time-factor sequence
implemented yesterday in TestChainLoad itself
- in this case, the TimeBase from the computation load is used as level speed
- this »base beat« is then modulated by the timing factor sequence
- working in an additional stress factor to press the schedule uniformly
- actual start time will be added as offset once the actual test commences
...up to now, we've relied on a regular schedule governed solely
by the progression of node levels, with a fixed level speed
defaulting to 1ms per level.
But in preparation of stress tesging, we need a schedule adapted
to the expected distribution of computation times, otherwise
we'll not be able to factor out the actual computation graph
connectivity. The goal is to establish a distinctive
**breaking point** when the scheduler is unable to cope with
the provided schedule.
The helper developed thus far produces a sequence of
weight factors per level, which could then be multiplied
with an actual delay base time to produce a concrete schedule.
These calculations, while simple, are difficult to understand;
recommended to use the values tabulated in this test together
with a `graphviz` rendering of the node graph (🠲 `printTopologyDOT()`)
The intention is to establish a theoretical limit for the expense,
given some degree of concurrency. In reality, the expense should always
be greater, since the time is not just split by the number of cores;
rather we need to chain up existing jobs of various weight on the available
cores (which is a special case of the box packing problem).
With this formula, an ideal weight factor can be determined for each level,
and then summing up the sequence of levels gives us a guess for a sensible
timing for the overall scheduler
...so IterExplorer got yet another processing layer,
which uses the grouping mechanics developed yesterday,
but is freely configurable through λ-Functions.
At actual usage sit in TestChainLoad, now only the actual
aggregation computation must be supplied, and follow-up computations
can now be chained up easily as further transformation layers.
Yesterday I've written a simple loop-based implementation of
a grouping aggregation to count the node weights per level.
Unfortunately it turns out we'll use several flavours of this
and we'd have to chain up postprocessing -- thus from a usage perspective
it would be better to have the same functionality packaged as interator pipeline.
This turns out to be surprisingly tricky and there is no suitable library
function available, which means I'll have to write one myself.
This changeset is the first step into this direction: reformulate
the simple for-loop into a demand-driven grouping iterator
...the idea is to use the sum of node weights per level
to create a schedule, which more closely reflects the distribution
of actual computation time. Hopefully such a schedule can then be
squeezed or stretched by a time factor to find out a ''breaking point'',
at which the Scheduler is no longer able to keep up.
In-depth investigation and reasoning highlighted another problem,
which could lead to memory corruption in rare cases; in the end
I found a solution by caching the ''address'' of the current Epoch
and re-validating this address on each Epoch-overflow.
After some difficulties getting any reliable measurement for a Release-build,
it turned out that this solution even ''improves performance by 22%''
Remark-1: the static blockFlow::Config prevents simple measurements by
just recompiling one translation unit; it is necessary to build the
relevant parts of Vault-layer with optimisation to get reliable numbers
Remark-2: performing a full non-DEBUG build highlighted two missing
header-inclusions to allow for the necessary template specialisations.
...discovered by during investigation of latest Scheduler failures.
The root of the problems is that block overflow can potentially trigger
expansion of the allocation pool. Under some circumstances, this on-the fly
allocation requires a rotation of index slots, thereby invalidating
existing iterators.
While such behaviour is not uncommon with storage data structures (see std::vector),
in this case it turns out problematic because due to performance considerations,
a usage pattern emerged which exploits re-using existing storage »Slots« with known
deadline. This optimisation seems to have significant leverage on the
planning jobs, which happen to allocated and arrange a whole strike of
Activities with similar deadlines.
One of these problem situations can easily be fixed, since it is triggered
through the iterator itself, using a delegate function to request a storage expansion,
at which point the iterator is able to re-link and fix its internal index.
This solution also has no tangible performance implications in optimised code.
Unfortunately there remains one obscure corner case where such an pool expansion
could also have invalidated other iterators, which are then used later to
attach dependency relations; even a partial fix for that problem seems
to cause considerable performance cost of about -14% in optimised code.
- now there can not be any direct dispatch anymore when entering events
- thus there is no decision logic at entrance anymore
- rather the work-function implementation moved down into Layer-2
- so add a unit-test like coverage there (integration in SchedulerService_test)
This amounts to a rather massive refactoring, prompted by the enduring problems
observed when pressing the scheduler. All the various glitches and (fixed) crashes
are related to the way how planning-jobs enter the schedule items,
which is also closely tied to the difficulties getting the locking
for planning-jobs correct.
The solution pursued hereby is to reorder the main avenues into the
scheduler implementation. There is now a streamlined main entrance,
which **always** enqueues only, allowing to omit most checks and
coordination. On the other hand, the complete coordination and dispatch
of the work capacity is now shifted down into the SchedulerCommutator,
thereby linking all coordination and access control close together
into a single implementation facility.
If this works out as intended
- several repeated checks on the Grooming-Token could be omitted (performance)
- the planning-job would no longer be able to loose / drop the Token,
thereby running enforcedly single-threaded (as was the original intention)
- since all planning effectively originates from planning-jobs, this
would allow to omit many safety barriers and complexities at the
scheduler entrance avenue, since now all entries just go into the queue.
WIP: tests pass compiler, but must be adapted / reworked
- fix mistake in schdule time for planning chunks (must use start, not end of chunk)
- allow to configure the heuristics for pre-roll (time reserved for planning a node)
...observing multiple failures, which seem to be interconnected
- the test-setup with the planning chunk pre-roll is insufficient
- basically it is not possible to perform further concurrent planning,
without getting behind the actual schedule; at least in the setup
with DUMP print statements (which slowdown everything)
- muliple chained re-entrant calls into the planning function can result
- the **ASSERTION in the Allocator** was triggered again
- the log+stacktrace indicate that there **is still a Gap**
in the logic to protect the allocations via Grooming-Token
...causing the system to freeze due to excess memory allocation.
Fortunately it turned out this was not an error in the Scheduler core
or memory manager, but rather a sloppiness in the test scaffolding.
However, this incident highlights that the memory manager lacks some
sanity checks to prevent outright nonsensical allocation requests.
Moreover it became clear again that the allocation happens ''already before''
entering the Scheduler — and thus the existing sanity check comes too late.
Now I've used the same reasoning also for additional checks in the allocator,
limiting the Epoch increment to 3000 and the total memory allocation to 8GiB
Talking of Gibitbytes...
indeed we could use a shorthand notation for that purpose...
while my basic assessment is still that contention will not play a significant
role given the expected real world usage scenario — when testing with
tighter schedule and rather short jobs (500µs), some phases of massive contention
can be observed, leading to significant slow-down of the test.
The major problem seems to be that extended phases of contention will
effectively cause several workers to remain in an active spinning-loop for
multiple microseconds, while also permanently reading the atomic lock.
Thus an adaptive scheme is introduced: after some repeated contention events,
workers now throttle down by themselves, with polling delays increased
with exponential stepping up to 2ms. This turns out to be surprisingly
effective and completely removes any observed delays in the test setup.
Turns out that we need to implemented fine grained and explicit handling logic
to ensure that Activity planning only ever happens protected by the Grooming-Token.
This is in accordance to the original design, which dictates that all management tasks
must be done in »management mode«, which can only be entered by a single thread at a time.
The underlying assumption is that the effort for management work is dwarfed in comparison
to any media calculation work.
However, in
5c6354882d
...I discovered an insidious border condition, an in an attempt to fix it,
I broke that fundamental assumpton. The problem arises from the fact that we
do want to expose a *public API* of the Scheduler. Even while this is only used
to ''seed'' a calculation stream, because any further planning- and management work
will be performed by the workers themselves (this is a design decision, we do not
employ a "scheduler thread")
Anyway, since the Scheduler API ''is'' public, ''someone from the outside'' could
invoke those functions, and — unaware of any Scheduler internals — will
automatically acquire the Grooming-Token, yet never release it,
leading to deadlock.
So we need a dedicated solution, which is hereby implemented as a
scoped guard: in the standard case, the caller is a management-job and
thus already holds the token (and nothing must be done). But in the
rare case of an »outsider«, this guard now ''transparently'' acquires
the token (possibly with a blocking wait) and ''drops it when leaving scope''
...this feature seems to be no longer necessary now;
leaving the actual implementation in-code for the time being,
but removed it from the public access API.
The last round of refactorings yielded significant improvements
- parallelisation now works as expected
- processing progresses closer to the schedule
- run time was reduced
The processing load for this test is tuned in a way to overload the
scheduler massively at the end -- the result must be correct non the less.
There was one notable glitch with an assertion failure from the memory manager.
Hopefully I can reproduce this by pressing and overloading the Scheduler more...
The rework from yesterday turned out to be effective ... unfortunately
a bit to much: since now late follow-up notifications take precedence,
a single worker tends to process the complete chain depth-first, because
the first chain will be followed and processed, even before the worker
was able to post the tasks for the other branches. Thus this single
worker is the only one to get a chance to proceed.
After some consideration, I am now leaning towards a fundamental change,
instead of just fixing some unfavourable behaviour pattern: while the
language semantics remains the same, the scheduler should no longer
directly dispatch into the next chain **from λ-post**. That is, whenever
a POST / NOTIFY is issued from the Activity-chain, the scheduler goes
through prioritisation.
This has further ramifications: we do not need a self-inhibition mechanism
any more (since now NOTIFY picks up the schedule time of the target).
With these changes, processing seems to proceed more smoothly,
albeit still with lots of contention on the Grooming token,
at least in the example structure tested here.
While the recent refactoring...
206c67cc
...was a step into the right direction, it pushed too hard,
overlooking the requirement to protect the scheduler contents
and thus all of the Activity-chains against concurrent modification.
Moreover, the recent solution still seems not quite orthogonal.
Thus the handling of notifications was thoroughly reworked:
- the explicit "double-dispatch" was removed, since actual usage
of the language indicates that we only need notifications to
Gate (and Hook), but not to any other conceivable Activity.
- thus it seems unnecessary to turn "notification" into some kind
of secondary work mode. Rather, it is folded as special case
into the regular dispatch.
This leads to new processing rules:
- a POST goes into λ-post (obviously... that's its meaning)
- a NOTIFY now passes its *target* into λ-post
- λ-post invokes ''dispatch''
- and **dispatching a Gate now implies to notify the Gate**
This greatly simplifies the »state machine« in the Activity-Language,
but also incurs some limitations (which seems adequate, since it is
now clear that we do not ''schedule'' or ''dispatch'' arbitrary
Activities — rather we'll do this only with POST and NOTIFY,
and all further processing happens by passing activation
along the chain, without involving the Scheduler)
use a feature of the Activity-Language prepared for this purpose:
self-Inhibition of the Chain. This prevents a prerequisite-NOTIFY
to trigger a complete chain of available tasks, before these tasks
have actually reached their nominal scheduling time.
This has the effect to align the computations much more strictly
with the defined schedule
The main (test) thread is kept in a blocking wait until the
planned schedule is completed. If however the schedule overruns,
the wake-up job could just be triggered prematurely.
This can easily be prevented by adding a dependency from the last
computation job to the wake-up job. If the computation somehow
flounders, the SAFETY_TIMEOUT (5s) will eventually raise
an exception to let the test fail cleanly (shutting down
the Scheduler automatically)
...it seems impossible to solve this conundrum other than by
opening a path to override a contextual deadline setting from
within the core Activity-Language logic.
This will be used in two cases
- when processing a explicitly coded POST (using deadline from the POST)
- after successfully opening a Gate by NOTIFY (using deadline from Gate)
All other cases can now supply Time::NEVER, thereby indicating that
the processing layer shall use contextual information (intersection
of the time intervals)
...this is an interesting test failure, which highlights inconsistencies
with handling of deadlines when processing follow-up from NOTIFY-triggers
There was also some fuzziness related to the ''meaning'' of λ-post,
leading to at least one superfluous POST invocation for each propagation;
fixing this does not solve the problem yet removes unnecessary overhead
and lock-contention
...playing around with the graph for the Scheduler integration test
...single threaded run time seemed to behave irregular
...but in fact it is very close to what can be expected
based on an ''averaged node weight''
Fortunately its very simple to add that into the existing node statistics
Basically this is all done and settled already: this is the `usageExample()`
from `TestChainLoadTest`. However, the focus is slightly different here:
We want a demonstration that the Scheduler can work flawlessly through
a massive load. Thus the plan is to use much more challenging parameters,
and then lean back and watch what happens....
...which turns out to be due to the DUMP-Statements,
which seem to create quite some contention on their own.
Test cases with very tight schedule will slip away then;
without print statement everything is GREEN now
This partially reverts commit 72f11549e6.
"Chain-Load: Scheduler instrumentation for observation"
Hint: revert this changeset to re-introduce the print statements for diagnostic
* added benchmark over synchronous execution as point of reference
* verified running times and execution pattern
* Scheduler **behaves as expected** for this example
- Generally speaking, the calibration uses current baseline settings;
- There are now two different load generation methods, thus both must be calibrated
- Performance contains some socked and non-linear effects, thus calibration
should be done close to the work point, which can be achieved by incremental
calibration until the error is < 5%
Interestingly, longer time-base values run slightly faster than predicted,
which is consistent with the expectation (socket cost). And using a larger
memory block increases time values, which is also plausible, since
cache effects will be diminishing
..initial gauging is a tricky subject,
since existing computer's performance spans a wide scale
Allowing
- pre calibration -98% .. +190%
- single run ±20%
- benchmark ±5%
...which can be deliberately attached (or not attached) to the
individual node invocation functor, allowing to study the effect
of actual load vs. zero-load and worker contention
Within Chain-Load, the infrastructure to add this crucial feature
is minimal: each node gets a `weight` parameter, which is assigned
using another RandomDraw-Rule (by default `weight==0`).
The actual computation load will be developed as a separate component
and tied in from the node calculation job functor.
...during development of the Chain-Load, it became clear that we'll often
need a collection of small trees rather than one huge graph. Thus a rule
for pruning nodes and finishing graphs was added. This has the consequence
that there might now be several exit nodes scattered all over the graph;
we still want one single global hash value to verify computations,
thus those exit hashes must now be picked up from the nodes and
combined into a single value.
All existing hash values hard coded into tests must be updated
...with this change, processing is ''ahead of schedule'' from the beginning,
which has the nice side effect that the problematic contention situation
with these very short computation jobs can not arise, and most of the schedule
is processed by a single worker.
Processing pattern is now pretty much as expected
This is a trick to get much better scheduling and timing guesses.
Instead of targeting a specific level, rather a fixed number of nodes
is processed in each chunk, yet still always processing complete levels.
The final level number to expect can be retrieved from the chain-load graph.
With this refactoring, we can now schedule a wake-up job precisely
after the expected completion of the last level
Scheduling a wake-up job behind the end of the planned schedule did the trick.
Sometimes there is ''strong contention'' immediately after full provision of the WorkForce,
but this seems to be as expected, since the »Jobs« currently used have no
actually relevant run time on their own. It is even more surprising that
the Capacity-control logic is able to cope with this situation in a matter
of just some milliseconds, bringing the average Lag at ~ 300µs
Invent a special JobFunctor...
- can be created / bound from a λ
- self-manages its storage on the heap
- can be invoked once, then discards itself
Intention is to pass such one-time actions to the Scheduler
to cause some ad-hoc transitions tied to curren circumstances;
a notable example will be the callback after load-test completion.
In the first draft version, a blocked Gate was handled by
»polling« the Gate regularly by scheduling a re-invocation
repeatedly into the future (by a stepping defined through
ExecutionCtx::getWaitDelay()).
Yet the further development of the Activity-Language indicates
that the ''Notification mechanism'' is sufficient to handle all
foreseeable aspects of dependency management. Consequently this
''Gate poling is no longer necessary,'' since on Notification
the Gate is automatically checked and the activation impulse
is immediately passed on; thus the re-scheduled check would
never get an opportunity actually to trigger the Gate; such
an active polling would only be necessary if the count down
latch in the Gate is changed by "external forces".
Moreover, the first Scheduler integration tests with TestChainLoad
indicate that the rescheduled polling can create a considerable
additional load when longer dependency chains miss one early
prerequisite, and this additional load (albeit processed
comparatively fast by the Scheduler) will be shifted along
needlessly for quite some time, until all of the activities
from the failed chain have passed their deadline. And what
is even more concerning, these useless checks have a tendency
to miss-focus the capacity management, as it seems there is
much work to do in a near horizon, which in fact may not be
the case altogether.
Thus the Gate implementation is now *changed to just SKIP*
when blocked. This helped to drastically improve the behaviour
of the Scheduler immediately after start-up -- further observation
indicated another adjustment: the first Tick-duty-cycle is now
shortened, because (after the additional "noise" from gate-rescheduling
was removed), the newly scaled-up work capacity has the tendency
to focus in the time horizon directly behind the first jobs added
to the timeline, which typically is now the first »Tick«.
ð¡ this leads to a recommendation, to arrange the first job-planning
chunk in such a way that the first actual work jobs appear in the area
between 5ms and 10ms after triggering the Scheduler start-up.Scheduler¡
The first complete integration test with Chain-Load
highlighted some difficulties with the overall load regulation:
- it works well in the standard case (but is possibly to eager to scale up)
- the scale-up sometimes needs several cycles to get "off the ground"
- when the first job is dispatched immediately instead of going
through the queue, the scheduler fails to boot up
two rather obvious bugfixes
(well, after watching the Scheduler in action...)
- the first planning-chunk needs an offset
- the future to block on must be setup before any dispatch happens
- prime diagnostics with the first time invocation
- print timings relative to this first invocation
- DUMP output to watch the crucial scheduling operations
... so this (finally) is the missing cornerstone
... traverse the calculation graph and generate render jobs
... provide a chunk-wise pre-planning of the next batch
... use a future to block the (test) thread until completed
- test setup without actual scheduler
- wire the callbacks such to verify
+ all nodes are touched
+ levels are processed to completion
+ the planning chunk stops at the expected level
+ all node dependencies are properly reported through the callbacks
- decided to abstract the scheduler invocations as λ
- so this functor contains the bare loop logic
Investigation regarding hash-framework:
It turns out that boost::hash uses a different hash_combine,
than what we have extracted/duplicated in lib/hash-value.hpp
(either this was a mistake, or boost::hash did use this weaker
function at that time and supplied a dedicated 64bit implementation later)
Anyway, should use boost::hash for the time being
maybe also fix the duplicated impl in lib/hash-value.hpp
- use a ''special encoding'' to marshal the specific coordinates for this test setup
- use a fixed Frame-Grid to represent the ''time level''
- invoke hash calculation through a specialised JobFunctor subclass
The number of nodes was just defined as template argument
to get a cheap implementation through std::array...
But actually this number of nodes is ''not a characteristics of the type;''
we'd end up with a distinct JobFunctor type for each different test size,
which is plain nonsensical. Usage analysis reveals, now that the implementation
is ''basically complete,'' that all of the topology generation and statistic
calculation code does not integrate deeply with the node storage, but
rather just iterates over all nodes and uses the ''first'' and ''last'' node.
This can actually be achieved very easy with a heap-allocated plain array,
relying on the magic of lib::IterExplorer for all iteration and transformation.
- use a dedicated context "dropped off" the TestChainLoad instance
- encode the node-idx into the InvocationInstanceID
- build an invocation- and a planning-job-functor
- let planning progress over an lib::UninitialisedStorage array
- plant the ActivityTerm instances into that array as Scheduling progresses
Since Chain-Load shall be used for performance testing of the scheduler,
we need a catalogue of realistic load patterns. This extended effort
started with some parameter configurations and developed various graph
shapes with different degree of connectivity and concurrency, ranging
from a stable sequence of very short chains to large and excessively
interconnected dependency networks.
Through introduction of a ''pruning rule'', it is possible
to create exit nodes in the middle of the graph. With increased
intensity of pruning, it is possible to ''choke off'' the generation
and terminate the graph; in such a case a new seed node is injected
automatically. By combination with seed rules, an equilibrium of
graph start and graph termination can be achieved.
Following this path, it should be possible to produce a pattern,
which is random but overall stable and well suited to simulate
a realistic processing load.
However, finding proper parameters turns out quite hard in practice,
since the behaviour is essentially contingent and most combinations
either lead to uninteresting trivial small graph chunks, or to
large, interconnected and exponentially expanding networks
... seeding happens at random points in the middle of the chain
... when combined with reduction, the resulting processing pattern
resembles the real processing pattern of media calcualtions
... special rule to generate a fixed expansion on each seed
... consecutive reductions join everything back into one chain
... can counterbalance expansions and reductions
...as it turns out, the solution embraced first was the cleanest way
to handle dynamic configuration of parameters; just it did not work
at that time, due to the reference binding problem in the Lambdas.
Meanwhile, the latter has been resolved by relying on the LazyInit
mechanism. Thus it is now possible to abandon the manipulation by
side effect and rather require the dynamic rule to return a
''pristine instance''.
With these adjustments, it is now possible to install a rule
which expands only for some kinds of nodes; this is used here
to crate a starting point for a **reduction rule** to kick in.
- present the weight centres relative to overall level count
- detect sub-graphs and add statistics per subgraph
- include an evaluation for ''all nodes''
- include number of levels and subgraphs
- iterate over all nodes and classify them
- group per level
- book in per level statistics into the Indicator records
- close global averages
...just coded, not yet tested...
The graph will be used to generate a computational load
for testing the Scheduler; thus we need to compute some
statistical indicators to characterise this load.
As starting point sum counts and averages will be aggregated,
accounting for particular characterisation of nodes per level.
It seams indicated to verify the generated connectivity
and the hash calculation and recalculation explicitly
at least for one example topology; choosing a topology
comprised of several sub-graphs, to also verify the
propagation of seed values to further start-nodes.
In order to avoid addressing nodes directly by index number,
those sub-graphs can be processed by ''grouping of nodes'';
all parts are congruent because topology is determined by
the node hashes and thus a regular pattern can be exploited.
To allow for easy processing of groups, I have developed a
simplistic grouping device within the IterExplorer framework.
- with the new pruning option, start-Nodes can now be anywhere
- introduce predicates to detect start-Nodes and exit-Nodes
- ensure each new seed node gets the global seed on graph construction
- provide functionality to re-propagate a seed and clear hashes
- provide functionality to recalculate the hashes over the graph
up to now, random values were completely determined by the
Node's hash, leading to completely symmetrical topology.
This is fine, but sometimes additional randomness is desirable,
while still keeping everything deterministic; the obvious solution
is to make the results optionally dependent on the invocation order,
which is simply to achieve with an additional state field. After some
tinkering, I decided to use the most simplistic solution, which is
just a multiplication with the state.
this is only a minor rearrangement in the Algorithm,
but allows to re-boot computation should node connectivity
go to zero. With current capabilities, this could not happen,
but I'm considering to add a »pruning« parameter to create the
possibility to generate multiple shorter chains instead of one
complete chain -- which more closely emulates reality for
Scheduler load patterns.
...so this was yet another digression, caused by the desire
somehow to salvage this problematic component design. Using a
DSL token fluently, while internally maintaining a complex and
totally open function based configuration is a bit of a stretch.
...this is a more realistic demo example, which mimics
some of the patterns present in RandomDraw. The test also
uses lambdas linking to the actual storage location, so that
the invocation would crash on a copy; LazyInit was invented
to safeguard against this, while still allowing leeway
during the initialisation phase in a DSL.
- Helper function to find out of two objects are located
"close to each other" -- which can be used as heuristics
to distinguish heap vs. stack storage
- further investigation shows that libstdc++ applies the
small-object optimisation for functor up to »two slots«
in size -- but only if the copy-ctor is trivial. Thus
a lambda capturing a shared_ptr by value will *always*
be maintained in heap storage (and LazyInit must be
redesigned accordingly)...
- the verify_inlineStorage() unit test will now trigger
if some implementation does not apply small-object optimisation
under these minimal assumptions
...which is crucial for the solution pursued at the moment;
std::function is known to apply a small-object optimisation,
yet unfortunately there are no guarantees by the C++ standard
(it is only mandated that std::function handles a bare function
pointer without overhead)
Other people have investigated that behaviour already,
indicating that at least one additional »slot« of data
can be handled with embedded storage in all known implementations
(while libstdc++ seemingly imposes the strongest limitations)
https://stackoverflow.com/a/77202545/444796
This experiment in the unit-test shows that for my setup
(libstdc++ and GCC-8) only a lambda capturing a single pointer
is handled entirely embedded into the std::function; already
a lambda capturing a shared-ptr leads to overflow into heap
