starpu-max/doc/doxygen/chapters/optimize_performance.doxy

/*
 * This file is part of the StarPU Handbook.
 * Copyright (C) 2009--2011  Universit@'e de Bordeaux 1
 * Copyright (C) 2010, 2011, 2012, 2013  Centre National de la Recherche Scientifique
 * Copyright (C) 2011, 2012 Institut National de Recherche en Informatique et Automatique
 * See the file version.doxy for copying conditions.
 */

/*! \page optimizePerformance How to optimize performance with StarPU

TODO: improve!

Simply encapsulating application kernels into tasks already permits to
seamlessly support CPU and GPUs at the same time. To achieve good performance, a
few additional changes are needed.

\section Data_management Data management

When the application allocates data, whenever possible it should use the
starpu_malloc() function, which will ask CUDA or
OpenCL to make the allocation itself and pin the corresponding allocated
memory. This is needed to permit asynchronous data transfer, i.e. permit data
transfer to overlap with computations. Otherwise, the trace will show that the
<c>DriverCopyAsync</c> state takes a lot of time, this is because CUDA or OpenCL
then reverts to synchronous transfers.

By default, StarPU leaves replicates of data wherever they were used, in case they
will be re-used by other tasks, thus saving the data transfer time. When some
task modifies some data, all the other replicates are invalidated, and only the
processing unit which ran that task will have a valid replicate of the data. If the application knows
that this data will not be re-used by further tasks, it should advise StarPU to
immediately replicate it to a desired list of memory nodes (given through a
bitmask). This can be understood like the write-through mode of CPU caches.

\code{.c}
starpu_data_set_wt_mask(img_handle, 1<<0);
\endcode

will for instance request to always automatically transfer a replicate into the
main memory (node 0), as bit 0 of the write-through bitmask is being set.

\code{.c}
starpu_data_set_wt_mask(img_handle, ~0U);
\endcode

will request to always automatically broadcast the updated data to all memory
nodes.

Setting the write-through mask to <c>~0U</c> can also be useful to make sure all
memory nodes always have a copy of the data, so that it is never evicted when
memory gets scarse.

Implicit data dependency computation can become expensive if a lot
of tasks access the same piece of data. If no dependency is required
on some piece of data (e.g. because it is only accessed in read-only
mode, or because write accesses are actually commutative), use the
starpu_data_set_sequential_consistency_flag() function to disable implicit
dependencies on that data.

In the same vein, accumulation of results in the same data can become a
bottleneck. The use of the <c>STARPU_REDUX</c> mode permits to optimize such
accumulation (see \ref Data_reduction).

Applications often need a data just for temporary results.  In such a case,
registration can be made without an initial value, for instance this produces a vector data:

\code{.c}
starpu_vector_data_register(&handle, -1, 0, n, sizeof(float));
\endcode

StarPU will then allocate the actual buffer only when it is actually needed,
e.g. directly on the GPU without allocating in main memory.

In the same vein, once the temporary results are not useful any more, the
data should be thrown away. If the handle is not to be reused, it can be
unregistered:

\code{.c}
starpu_unregister_submit(handle);
\endcode

actual unregistration will be done after all tasks working on the handle
terminate.

If the handle is to be reused, instead of unregistering it, it can simply be invalidated:

\code{.c}
starpu_invalidate_submit(handle);
\endcode

the buffers containing the current value will then be freed, and reallocated
only when another task writes some value to the handle.

\section Task_granularity Task granularity

Like any other runtime, StarPU has some overhead to manage tasks. Since
it does smart scheduling and data management, that overhead is not always
neglectable. The order of magnitude of the overhead is typically a couple of
microseconds, which is actually quite smaller than the CUDA overhead itself. The
amount of work that a task should do should thus be somewhat
bigger, to make sure that the overhead becomes neglectible. The offline
performance feedback can provide a measure of task length, which should thus be
checked if bad performance are observed. To get a grasp at the scalability
possibility according to task size, one can run
<c>tests/microbenchs/tasks_size_overhead.sh</c> which draws curves of the
speedup of independent tasks of very small sizes.

The choice of scheduler also has impact over the overhead: for instance, the
<c>dmda</c> scheduler takes time to make a decision, while <c>eager</c> does
not. <c>tasks_size_overhead.sh</c> can again be used to get a grasp at how much
impact that has on the target machine.

\section Task_submission Task submission

To let StarPU make online optimizations, tasks should be submitted
asynchronously as much as possible. Ideally, all the tasks should be
submitted, and mere calls to starpu_task_wait_for_all() or
starpu_data_unregister() be done to wait for
termination. StarPU will then be able to rework the whole schedule, overlap
computation with communication, manage accelerator local memory usage, etc.

\section Task_priorities Task priorities

By default, StarPU will consider the tasks in the order they are submitted by
the application. If the application programmer knows that some tasks should
be performed in priority (for instance because their output is needed by many
other tasks and may thus be a bottleneck if not executed early enough), the
<c>priority</c> field of the task structure should be set to transmit the
priority information to StarPU.

\section Task_scheduling_policy Task scheduling policy

By default, StarPU uses the <c>eager</c> simple greedy scheduler. This is
because it provides correct load balance even if the application codelets do not
have performance models. If your application codelets have performance models
(\ref Performance_model_example for example showing how to do it),
you should change the scheduler thanks to the <c>STARPU_SCHED</c> environment
variable. For instance <c>export STARPU_SCHED=dmda</c> . Use <c>help</c> to get
the list of available schedulers.

The <b>eager</b> scheduler uses a central task queue, from which workers draw tasks
to work on. This however does not permit to prefetch data since the scheduling
decision is taken late. If a task has a non-0 priority, it is put at the front of the queue.

The <b>prio</b> scheduler also uses a central task queue, but sorts tasks by
priority (between -5 and 5).

The <b>random</b> scheduler distributes tasks randomly according to assumed worker
overall performance.

The <b>ws</b> (work stealing) scheduler schedules tasks on the local worker by
default. When a worker becomes idle, it steals a task from the most loaded
worker.

The <b>dm</b> (deque model) scheduler uses task execution performance models into account to
perform an HEFT-similar scheduling strategy: it schedules tasks where their
termination time will be minimal.

The <b>dmda</b> (deque model data aware) scheduler is similar to dm, it also takes
into account data transfer time.

The <b>dmdar</b> (deque model data aware ready) scheduler is similar to dmda,
it also sorts tasks on per-worker queues by number of already-available data
buffers.

The <b>dmdas</b> (deque model data aware sorted) scheduler is similar to dmda, it
also supports arbitrary priority values.

The <b>heft</b> (heterogeneous earliest finish time) scheduler is deprecated. It
is now just an alias for <b>dmda</b>.

The <b>pheft</b> (parallel HEFT) scheduler is similar to heft, it also supports
parallel tasks (still experimental).

The <b>peager</b> (parallel eager) scheduler is similar to eager, it also
supports parallel tasks (still experimental).

\section Performance_model_calibration Performance model calibration

Most schedulers are based on an estimation of codelet duration on each kind
of processing unit. For this to be possible, the application programmer needs
to configure a performance model for the codelets of the application (see
\ref Performance_model_example for instance). History-based performance models
use on-line calibration.  StarPU will automatically calibrate codelets
which have never been calibrated yet, and save the result in
<c>$STARPU_HOME/.starpu/sampling/codelets</c>.
The models are indexed by machine name. To share the models between machines (e.g. for a homogeneous cluster), use <c>export STARPU_HOSTNAME=some_global_name</c>. To force continuing calibration, use
<c>export STARPU_CALIBRATE=1</c> . This may be necessary if your application
has not-so-stable performance. StarPU will force calibration (and thus ignore
the current result) until 10 (_STARPU_CALIBRATION_MINIMUM) measurements have been
made on each architecture, to avoid badly scheduling tasks just because the
first measurements were not so good. Details on the current performance model status
can be obtained from the <c>starpu_perfmodel_display</c> command: the <c>-l</c>
option lists the available performance models, and the <c>-s</c> option permits
to choose the performance model to be displayed. The result looks like:

\verbatim
$ starpu_perfmodel_display -s starpu_dlu_lu_model_22
performance model for cpu
# hash    size     mean          dev           n
880805ba  98304    2.731309e+02  6.010210e+01  1240
b50b6605  393216   1.469926e+03  1.088828e+02  1240
5c6c3401  1572864  1.125983e+04  3.265296e+03  1240
\endverbatim

Which shows that for the LU 22 kernel with a 1.5MiB matrix, the average
execution time on CPUs was about 11ms, with a 3ms standard deviation, over
1240 samples. It is a good idea to check this before doing actual performance
measurements.

A graph can be drawn by using the <c>starpu_perfmodel_plot</c>:

\verbatim
$ starpu_perfmodel_plot -s starpu_dlu_lu_model_22
98304 393216 1572864
$ gnuplot starpu_starpu_dlu_lu_model_22.gp
$ gv starpu_starpu_dlu_lu_model_22.eps
\endverbatim

If a kernel source code was modified (e.g. performance improvement), the
calibration information is stale and should be dropped, to re-calibrate from
start. This can be done by using <c>export STARPU_CALIBRATE=2</c>.

Note: due to CUDA limitations, to be able to measure kernel duration,
calibration mode needs to disable asynchronous data transfers. Calibration thus
disables data transfer / computation overlapping, and should thus not be used
for eventual benchmarks. Note 2: history-based performance models get calibrated
only if a performance-model-based scheduler is chosen.

The history-based performance models can also be explicitly filled by the
application without execution, if e.g. the application already has a series of
measurements. This can be done by using starpu_perfmodel_update_history(),
for instance:

\code{.c}
static struct starpu_perfmodel perf_model = {
    .type = STARPU_HISTORY_BASED,
    .symbol = "my_perfmodel",
};

struct starpu_codelet cl = {
    .where = STARPU_CUDA,
    .cuda_funcs = { cuda_func1, cuda_func2, NULL },
    .nbuffers = 1,
    .modes = {STARPU_W},
    .model = &perf_model
};

void feed(void) {
    struct my_measure *measure;
    struct starpu_task task;
    starpu_task_init(&task);

    task.cl = &cl;

    for (measure = &measures[0]; measure < measures[last]; measure++) {
        starpu_data_handle_t handle;
	starpu_vector_data_register(&handle, -1, 0, measure->size, sizeof(float));
	task.handles[0] = handle;
	starpu_perfmodel_update_history(&perf_model, &task,
	                                STARPU_CUDA_DEFAULT + measure->cudadev, 0,
	                                measure->implementation, measure->time);
	starpu_task_clean(&task);
	starpu_data_unregister(handle);
    }
}
\endcode

Measurement has to be provided in milliseconds for the completion time models,
and in Joules for the energy consumption models.

\section Task_distribution_vs_Data_transfer Task distribution vs Data transfer

Distributing tasks to balance the load induces data transfer penalty. StarPU
thus needs to find a balance between both. The target function that the
<c>dmda</c> scheduler of StarPU
tries to minimize is <c>alpha * T_execution + beta * T_data_transfer</c>, where
<c>T_execution</c> is the estimated execution time of the codelet (usually
accurate), and <c>T_data_transfer</c> is the estimated data transfer time. The
latter is estimated based on bus calibration before execution start,
i.e. with an idle machine, thus without contention. You can force bus re-calibration by running
<c>starpu_calibrate_bus</c>. The beta parameter defaults to 1, but it can be
worth trying to tweak it by using <c>export STARPU_SCHED_BETA=2</c> for instance,
since during real application execution, contention makes transfer times bigger.
This is of course imprecise, but in practice, a rough estimation already gives
the good results that a precise estimation would give.

\section Data_prefetch Data prefetch

The <c>heft</c>, <c>dmda</c> and <c>pheft</c> scheduling policies perform data prefetch (see @ref{STARPU_PREFETCH}):
as soon as a scheduling decision is taken for a task, requests are issued to
transfer its required data to the target processing unit, if needeed, so that
when the processing unit actually starts the task, its data will hopefully be
already available and it will not have to wait for the transfer to finish.

The application may want to perform some manual prefetching, for several reasons
such as excluding initial data transfers from performance measurements, or
setting up an initial statically-computed data distribution on the machine
before submitting tasks, which will thus guide StarPU toward an initial task
distribution (since StarPU will try to avoid further transfers).

This can be achieved by giving the starpu_data_prefetch_on_node() function
the handle and the desired target memory node.

\section Power-based_scheduling Power-based scheduling

If the application can provide some power performance model (through
the <c>power_model</c> field of the codelet structure), StarPU will
take it into account when distributing tasks. The target function that
the <c>dmda</c> scheduler minimizes becomes <c>alpha * T_execution +
beta * T_data_transfer + gamma * Consumption</c> , where <c>Consumption</c>
is the estimated task consumption in Joules. To tune this parameter, use
<c>export STARPU_SCHED_GAMMA=3000</c> for instance, to express that each Joule
(i.e kW during 1000us) is worth 3000us execution time penalty. Setting
<c>alpha</c> and <c>beta</c> to zero permits to only take into account power consumption.

This is however not sufficient to correctly optimize power: the scheduler would
simply tend to run all computations on the most energy-conservative processing
unit. To account for the consumption of the whole machine (including idle
processing units), the idle power of the machine should be given by setting
<c>export STARPU_IDLE_POWER=200</c> for 200W, for instance. This value can often
be obtained from the machine power supplier.

The power actually consumed by the total execution can be displayed by setting
<c>export STARPU_PROFILING=1 STARPU_WORKER_STATS=1</c> .

On-line task consumption measurement is currently only supported through the
<c>CL_PROFILING_POWER_CONSUMED</c> OpenCL extension, implemented in the MoviSim
simulator. Applications can however provide explicit measurements by using the
starpu_perfmodel_update_history() function (examplified in \ref Performance_model_example
with the <c>power_model</c> performance model. Fine-grain
measurement is often not feasible with the feedback provided by the hardware, so
the user can for instance run a given task a thousand times, measure the global
consumption for that series of tasks, divide it by a thousand, repeat for
varying kinds of tasks and task sizes, and eventually feed StarPU
with these manual measurements through starpu_perfmodel_update_history().

\section Static_scheduling Static scheduling

In some cases, one may want to force some scheduling, for instance force a given
set of tasks to GPU0, another set to GPU1, etc. while letting some other tasks
be scheduled on any other device. This can indeed be useful to guide StarPU into
some work distribution, while still letting some degree of dynamism. For
instance, to force execution of a task on CUDA0:

\code{.c}
task->execute_on_a_specific_worker = 1;
task->worker = starpu_worker_get_by_type(STARPU_CUDA_WORKER, 0);
\endcode

\section Profiling Profiling

A quick view of how many tasks each worker has executed can be obtained by setting
<c>export STARPU_WORKER_STATS=1</c> This is a convenient way to check that
execution did happen on accelerators without penalizing performance with
the profiling overhead.

A quick view of how much data transfers have been issued can be obtained by setting
<c>export STARPU_BUS_STATS=1</c> .

More detailed profiling information can be enabled by using <c>export STARPU_PROFILING=1</c> or by
calling starpu_profiling_status_set() from the source code.
Statistics on the execution can then be obtained by using <c>export
STARPU_BUS_STATS=1</c> and <c>export STARPU_WORKER_STATS=1</c> .
 More details on performance feedback are provided by the next chapter.

\section CUDA-specific_optimizations CUDA-specific optimizations

Due to CUDA limitations, StarPU will have a hard time overlapping its own
communications and the codelet computations if the application does not use a
dedicated CUDA stream for its computations instead of the default stream,
which synchronizes all operations of the GPU. StarPU provides one by the use
of starpu_cuda_get_local_stream() which can be used by all CUDA codelet
operations to avoid this issue. For instance:

\code{.c}
func <<<grid,block,0,starpu_cuda_get_local_stream()>>> (foo, bar);
cudaStreamSynchronize(starpu_cuda_get_local_stream());
\endcode

StarPU already does appropriate calls for the CUBLAS library.

Unfortunately, some CUDA libraries do not have stream variants of
kernels. That will lower the potential for overlapping.

\section Performance_debugging Performance debugging

To get an idea of what is happening, a lot of performance feedback is available,
detailed in the next chapter. The various informations should be checked for.

<ul>
<li>
What does the Gantt diagram look like? (see \ref Creating_a_Gantt_Diagram)
<ul>
  <li> If it's mostly green (tasks running in the initial context) or context specific 
  color prevailing, then the machine is properly
  utilized, and perhaps the codelets are just slow. Check their performance, see
  \ref Performance_of_codelets.
  </li>
  <li> If it's mostly purple (FetchingInput), tasks keep waiting for data
  transfers, do you perhaps have far more communication than computation? Did
  you properly use CUDA streams to make sure communication can be
  overlapped? Did you use data-locality aware schedulers to avoid transfers as
  much as possible?
  </li>
  <li> If it's mostly red (Blocked), tasks keep waiting for dependencies,
  do you have enough parallelism? It might be a good idea to check what the DAG
  looks like (see \ref Creating_a_DAG_with_graphviz).
  </li>
  <li> If only some workers are completely red (Blocked), for some reason the
  scheduler didn't assign tasks to them. Perhaps the performance model is bogus,
  check it (see \ref Performance_of_codelets). Do all your codelets have a
  performance model?  When some of them don't, the schedulers switches to a
  greedy algorithm which thus performs badly.
  </li>
</ul>
</li>
</ul>

You can also use the Temanejo task debugger (see \ref Using_the_Temanejo_task_debugger) to
visualize the task graph more easily.

\section Simulated_performance Simulated performance

StarPU can use Simgrid in order to simulate execution on an arbitrary
platform.

\subsection Calibration Calibration

The idea is to first compile StarPU normally, and run the application,
so as to automatically benchmark the bus and the codelets.

\verbatim
$ ./configure && make
$ STARPU_SCHED=dmda ./examples/matvecmult/matvecmult
[starpu][_starpu_load_history_based_model] Warning: model matvecmult
   is not calibrated, forcing calibration for this run. Use the
   STARPU_CALIBRATE environment variable to control this.
$ ...
$ STARPU_SCHED=dmda ./examples/matvecmult/matvecmult
TEST PASSED
\endverbatim

Note that we force to use the dmda scheduler to generate performance
models for the application. The application may need to be run several
times before the model is calibrated.

\subsection Simulation Simulation

Then, recompile StarPU, passing <c>--enable-simgrid</c> to <c>./configure</c>, and re-run the
application:

\verbatim
$ ./configure --enable-simgrid && make
$ STARPU_SCHED=dmda ./examples/matvecmult/matvecmult
TEST FAILED !!!
\endverbatim

It is normal that the test fails: since the computation are not actually done
(that is the whole point of simgrid), the result is wrong, of course.

If the performance model is not calibrated enough, the following error
message will be displayed

\verbatim
$ STARPU_SCHED=dmda ./examples/matvecmult/matvecmult
[starpu][_starpu_load_history_based_model] Warning: model matvecmult
    is not calibrated, forcing calibration for this run. Use the
    STARPU_CALIBRATE environment variable to control this.
[starpu][_starpu_simgrid_execute_job][assert failure] Codelet
    matvecmult does not have a perfmodel, or is not calibrated enough
\endverbatim

The number of devices can be chosen as usual with <c>STARPU_NCPU</c>,
<c>STARPU_NCUDA</c>, and <c>STARPU_NOPENCL</c>.  For now, only the number of
cpus can be arbitrarily chosen. The number of CUDA and OpenCL devices have to be
lower than the real number on the current machine.

The amount of simulated GPU memory is for now unbound by default, but
it can be chosen by hand through the <c>STARPU_LIMIT_CUDA_MEM</c>,
<c>STARPU_LIMIT_CUDA_devid_MEM</c>, <c>STARPU_LIMIT_OPENCL_MEM</c>, and
<c>STARPU_LIMIT_OPENCL_devid_MEM</c> environment variables.

The Simgrid default stack size is small; to increase it use the
parameter <c>--cfg=contexts/stack_size</c>, for example:

\verbatim
$ ./example --cfg=contexts/stack_size:8192
TEST FAILED !!!
\endverbatim

Note: of course, if the application uses <c>gettimeofday</c> to make its
performance measurements, the real time will be used, which will be bogus. To
get the simulated time, it has to use starpu_timing_now() which returns the
virtual timestamp in ms.

\subsection Simulation_on_another_machine Simulation on another machine

The simgrid support even permits to perform simulations on another machine, your
desktop, typically. To achieve this, one still needs to perform the Calibration
step on the actual machine to be simulated, then copy them to your desktop
machine (the <c>$STARPU_HOME/.starpu</c> directory). One can then perform the
Simulation step on the desktop machine, by setting the <c>STARPU_HOSTNAME</c>
environment variable to the name of the actual machine, to make StarPU use the
performance models of the simulated machine even on the desktop machine.

If the desktop machine does not have CUDA or OpenCL, StarPU is still able to
use simgrid to simulate execution with CUDA/OpenCL devices, but the application
source code will probably disable the CUDA and OpenCL codelets in that
case. Since during simgrid execution, the functions of the codelet are actually
not called, one can use dummy functions such as the following to still permit
CUDA or OpenCL execution:

\code{.c}
static struct starpu_codelet cl11 =
{
	.cpu_funcs = {chol_cpu_codelet_update_u11, NULL},
#ifdef STARPU_USE_CUDA
	.cuda_funcs = {chol_cublas_codelet_update_u11, NULL},
#elif defined(STARPU_SIMGRID)
	.cuda_funcs = {(void*)1, NULL},
#endif
	.nbuffers = 1,
	.modes = {STARPU_RW},
	.model = &chol_model_11
};
\endcode

*/