/* StarPU --- Runtime system for heterogeneous multicore architectures.
*
* Copyright (C) 2013-2020 Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
* Copyright (C) 2013 Simon Archipoff
*
* StarPU is free software; you can redistribute it and/or modify
* it under the terms of the GNU Lesser General Public License as published by
* the Free Software Foundation; either version 2.1 of the License, or (at
* your option) any later version.
*
* StarPU is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
*
* See the GNU Lesser General Public License in COPYING.LGPL for more details.
*/
/*! \page HowToDefineANewSchedulingPolicy How To Define A New Scheduling Policy
\section NewSchedulingPolicy_Introduction Introduction
StarPU provides two ways of defining a scheduling policy, a basic monolithic
way, and a modular way.
The basic monolithic way is directly connected with the core of StarPU, which
means that the policy then has to handle all performance details, such as data
prefetching, task performance model calibration, worker locking, etc.
examples/scheduler/dummy_sched.c is a trivial example which does not
handle this, and thus e.g. does not achieve any data prefetching or smart
scheduling.
The modular way allows to implement just one component, and
reuse existing components to cope with all these details.
examples/scheduler/dummy_modular_sched.c is a trivial example very
similar to dummy_sched.c, but implemented as a component, which allows to
assemble it with other components, and notably get data prefetching support for
free, and task performance model calibration is properly performed, which allows
to easily extend it into taking task duration into account, etc.
\section SchedulingHelpers Helper functions for defining a scheduling policy (Basic or modular)
Make sure to have a look at the \ref API_Scheduling_Policy section, which
provides a complete list of the functions available for writing advanced schedulers.
This includes getting an estimation for a task computation completion with
starpu_task_expected_length(), for the required data transfers with
starpu_task_expected_data_transfer_time_for(), for the required energy with
starpu_task_expected_energy(), etc. Per-worker variants are also available with
starpu_task_worker_expected_length(), etc. Other
useful functions include starpu_transfer_bandwidth(), starpu_transfer_latency(),
starpu_transfer_predict(), ...
One can also directly test the presence of a data handle with starpu_data_is_on_node().
Prefetches can be triggered by calling either starpu_prefetch_task_input_for(),
starpu_idle_prefetch_task_input(), starpu_prefetch_task_input_for_prio(), or
starpu_idle_prefetch_task_input_for_prio(). The _prio versions allow to
specify a priority for the transfer (instead of taking the task priority by
default). These prefetches are only processed when there are no fetch data
requests (i.e. a task is waiting for it) to process. The _idle versions
queue the transfers on the idle prefetch queue, which is only processed when
there are no non-idle prefetch to process.
starpu_get_prefetch_flag() is a convenient helper for checking the value of the
\ref STARPU_PREFETCH environment variable.
Usual functions can be used on tasks, for instance one can use the following to
get the data size for a task.
\code{.c}
size = 0;
write = 0;
if (task->cl)
for (i = 0; i < STARPU_TASK_GET_NBUFFERS(task); i++)
{
starpu_data_handle_t data = STARPU_TASK_GET_HANDLE(task, i)
size_t datasize = starpu_data_get_size(data);
size += datasize;
if (STARPU_TASK_GET_MODE(task, i) & STARPU_W)
write += datasize;
}
\endcode
Task queues can be implemented with the starpu_task_list functions.
Access to the \c hwloc topology is available with starpu_worker_get_hwloc_obj().
\section DefiningANewBasicSchedulingPolicy Defining A New Basic Scheduling Policy
A full example showing how to define a new scheduling policy is available in
the StarPU sources in examples/scheduler/dummy_sched.c.
The scheduler has to provide methods:
\code{.c}
static struct starpu_sched_policy dummy_sched_policy =
{
.init_sched = init_dummy_sched,
.deinit_sched = deinit_dummy_sched,
.add_workers = dummy_sched_add_workers,
.remove_workers = dummy_sched_remove_workers,
.push_task = push_task_dummy,
.pop_task = pop_task_dummy,
.policy_name = "dummy",
.policy_description = "dummy scheduling strategy"
};
\endcode
The idea is that when a task becomes ready for execution, the
starpu_sched_policy::push_task method is called to give the ready task to the
scheduler. When a worker is idle, the starpu_sched_policy::pop_task method is
called to get a task from the scheduler. It is up to the
scheduler to implement what is between. A simple eager scheduler is for instance
to make starpu_sched_policy::push_task push the task to a global list, and make
starpu_sched_policy::pop_task pop from this list. A scheduler can also use
starpu_push_local_task() to directly push tasks to a per-worker queue, and then
starpu does not even need to implement starpu_sched_policy::pop_task.
If there are no ready tasks within the scheduler, it can just return \c NULL, and
the worker will sleep.
The \ref starpu_sched_policy section provides the exact rules that govern the
methods of the policy.
One can enumerate the workers with this iterator:
\code{.c}
struct starpu_worker_collection *workers = starpu_sched_ctx_get_worker_collection(sched_ctx_id);
struct starpu_sched_ctx_iterator it;
workers->init_iterator(workers, &it);
while(workers->has_next(workers, &it))
{
unsigned worker = workers->get_next(workers, &it);
...
}
\endcode
To provide synchronization between workers, a per-worker lock exists to protect
the data structures of a given worker. It is acquired around scheduler methods,
so that the scheduler does not need any additional mutex to protect its per-worker data.
In case the scheduler wants to access another scheduler's data, it should use
starpu_worker_lock() and starpu_worker_unlock().
Calling \code{.c}starpu_worker_lock(B)\endcode from a worker \c A will however thus make
worker \c A wait for worker \c B to complete its scheduling method. That may be
a problem if that method takes a long time, because it is e.g. computing a
heuristic or waiting for another mutex, or even cause deadlocks if worker \c B is
calling \code{.c}starpu_worker_lock(A)\endcode at the same time. In such a case, worker \c B must
call starpu_worker_relax_on() and starpu_worker_relax_off() around the section
which potentially blocks (and does not actually need protection). While a worker
is in relaxed mode, e.g. between a pair of starpu_worker_relax_on() and
starpu_worker_relax_off() calls, its state can be altered by other threads: for
instance, worker \c A can push tasks for worker \c B. In consequence, worker \c B
must re-assess its state after \code{.c}starpu_worker_relax_off(B)\endcode, such as taking
possible new tasks pushed to its queue into account.
When the starpu_sched_policy::push_task method has pushed a task for another
worker, one has to call starpu_wake_worker_relax_light() so that the worker wakes up
and picks it. If the task was pushed on a shared queue, one may want to only
wake one idle worker. An example doing this is available in
src/sched_policies/eager_central_policy.c.
A pointer to one data structure specific to the scheduler can be set with
starpu_sched_ctx_set_policy_data() and fetched with
starpu_sched_ctx_get_policy_data(). Per-worker data structures can then be
store in it by allocating a \ref STARPU_NMAXWORKERS -sized array of structures indexed
by workers.
A variety of examples of
advanced schedulers can be read in src/sched_policies, for
instance random_policy.c, eager_central_policy.c,
work_stealing_policy.c Code protected by
if (_starpu_get_nsched_ctxs() > 1) can be ignored, this is for scheduling
contexts, which is an experimental feature.
\section DefiningANewModularSchedulingPolicy Defining A New Modular Scheduling Policy
StarPU's Modularized Schedulers are made of individual Scheduling Components
Modularizedly assembled as a Scheduling Tree. Each Scheduling Component has an
unique purpose, such as prioritizing tasks or mapping tasks over resources.
A typical Scheduling Tree is shown below.
|
starpu_push_task |
|
v
Fifo_Component
| ^
Push | | Can_Push
v |
Eager_Component
| ^
| |
v |
--------><-------------------><---------
| ^ | ^
Push | | Can_Push Push | | Can_Push
v | v |
Fifo_Component Fifo_Component
| ^ | ^
Pull | | Can_Pull Pull | | Can_Pull
v | v |
Worker_Component Worker_Component
| |
starpu_pop_task | |
v v
When a task is pushed by StarPU in a Modularized Scheduler, the task moves from
a Scheduling Component to an other, following the hierarchy of the
Scheduling Tree, and is stored in one of the Scheduling Components of the
strategy.
When a worker wants to pop a task from the Modularized Scheduler, the
corresponding Worker Component of the Scheduling Tree tries to pull a task from
its parents, following the hierarchy, and gives it to the worker if it succeded
to get one.
\subsection Interface
Each Scheduling Component must follow the following pre-defined Interface
to be able to interact with other Scheduling Components.
- push_task (child_component, Task) \n
The calling Scheduling Component transfers a task to its
Child Component. When the Push function returns, the task no longer
belongs to the calling Component. The Modularized Schedulers'
model relies on this function to perform prefetching.
See starpu_sched_component::push_task for more details
- pull_task (parent_component, caller_component) -> Task \n
The calling Scheduling Component requests a task from
its Parent Component. When the Pull function ends, the returned
task belongs to the calling Component.
See starpu_sched_component::pull_task for more details
- can_push (caller_component, parent_component) \n
The calling Scheduling Component notifies its Parent Component that
it is ready to accept new tasks.
See starpu_sched_component::can_push for more details
- can_pull (caller_component, child_component) \n
The calling Scheduling Component notifies its Child Component
that it is ready to give new tasks.
See starpu_sched_component::can_pull for more details
The components also provide the following useful methods:
- starpu_sched_component::estimated_load provides an estimated load of
the component
- starpu_sched_component::estimated_end provides an estimated date of
availability of workers behind the component, after processing tasks in
the component and below.
This is computed only if the estimated field of the tasks have been set
before passing it to the component.
\subsection BuildAModularizedScheduler Building a Modularized Scheduler
\subsubsection PreImplementedComponents Pre-implemented Components
StarPU is currently shipped with the following four Scheduling Components :
- Storage Components : Fifo, Prio \n
Components which store tasks. They can also prioritize them if
they have a defined priority. It is possible to define a threshold
for those Components following two criterias : the number of tasks
stored in the Component, or the sum of the expected length of all
tasks stored in the Component. When a push operation tries to queue a
task beyond the threshold, the push fails. When some task leaves the
queue (and thus possibly more tasks can fit), this component calls
can_push from ancestors.
- Resource-Mapping Components : Mct, Heft, Eager, Random, Work-Stealing \n
"Core" of the Scheduling Strategy, those Components are the
ones who make scheduling choices between their children components.
- Worker Components : Worker \n
Each Worker Component modelizes a concrete worker, and copes with the
technical tricks of interacting with the StarPU core. Modular schedulers
thus usually have them at the bottom of their component tree.
- Special-Purpose Components : Perfmodel_Select, Best_Implementation \n
Components dedicated to original purposes. The Perfmodel_Select
Component decides which Resource-Mapping Component should be used to
schedule a task: a component that assumes tasks with a calibrated
performance model; a component for non-yet-calibrated tasks, that will
distribute them to get measurements done as quickly as possible; and a
component that takes the tasks without performance models.\n
The Best_Implementation Component chooses which
implementation of a task should be used on the chosen resource.
\subsubsection ProgressionAndValidationRules Progression And Validation Rules
Some rules must be followed to ensure the correctness of a Modularized
Scheduler :
- At least one Storage Component without threshold is needed in a
Modularized Scheduler, to store incoming tasks from StarPU. It can for
instance be a global component at the top of the tree, or one component
per worker at the bottom of the tree, or intermediate assemblies. The
important point is that the starpu_sched_component::push_task call at the top can not
fail, so there has to be a storage component without threshold between
the top of the tree and the first storage component with threshold, or
the workers themselves.
- At least one Resource-Mapping Component is needed in a Modularized
Scheduler. Resource-Mapping Components are the only ones which can make
scheduling choices, and so the only ones which can have several child.
\subsubsection ModularizedSchedulerLocking Locking in modularized schedulers
Most often, components do not need to take locks. This allows e.g. the push
operation to be called in parallel when tasks get released in parallel from
different workers which have completed different ancestor tasks.
When a component has internal information which needs to be kept coherent, the
component can define its own lock at take it as it sees fit, e.g. to protect a
task queue. This may however limit scalability of the scheduler. Conversely,
since push and pull operations will be called concurrently from different
workers, the component might prefer to use a central mutex to serialize all
scheduling decisions to avoid pathological cases (all push calls decide to put
their task on the same target)
\subsubsection ImplementAModularizedScheduler Implementing a Modularized Scheduler
The following code shows how to implement a Tree-Eager-Prefetching Scheduler.
\code{.c}
static void initialize_eager_prefetching_center_policy(unsigned sched_ctx_id)
{
/* The eager component will decide for each task which worker will run it,
* and we want fifos both above and below the component */
starpu_sched_component_initialize_simple_scheduler(
starpu_sched_component_eager_create, NULL,
STARPU_SCHED_SIMPLE_DECIDE_WORKERS |
STARPU_SCHED_SIMPLE_FIFO_ABOVE |
STARPU_SCHED_SIMPLE_FIFOS_BELOW,
sched_ctx_id);
}
/* Properly destroy the Scheduling Tree and all its Components */
static void deinitialize_eager_prefetching_center_policy(unsigned sched_ctx_id)
{
struct starpu_sched_tree * tree = (struct starpu_sched_tree*)starpu_sched_ctx_get_policy_data(sched_ctx_id);
starpu_sched_tree_destroy(tree);
}
/* Initializing the starpu_sched_policy struct associated to the Modularized
* Scheduler : only the init_sched and deinit_sched needs to be defined to
* implement a Modularized Scheduler */
struct starpu_sched_policy _starpu_sched_tree_eager_prefetching_policy =
{
.init_sched = initialize_eager_prefetching_center_policy,
.deinit_sched = deinitialize_eager_prefetching_center_policy,
.add_workers = starpu_sched_tree_add_workers,
.remove_workers = starpu_sched_tree_remove_workers,
.push_task = starpu_sched_tree_push_task,
.pop_task = starpu_sched_tree_pop_task,
.pre_exec_hook = starpu_sched_component_worker_pre_exec_hook,
.post_exec_hook = starpu_sched_component_worker_post_exec_hook,
.pop_every_task = NULL,
.policy_name = "tree-eager-prefetching",
.policy_description = "eager with prefetching tree policy"
};
\endcode
starpu_sched_component_initialize_simple_scheduler() is a helper function which
makes it very trivial to assemble a modular scheduler around a scheduling
decision component as seen above (here, a dumb eager decision component). Most
often a modular scheduler can be implemented that way.
A modular scheduler can also be constructed hierarchically with
starpu_sched_component_composed_recipe_create().
That modular scheduler can also be built by hand in the following way:
\code{.c}
#define _STARPU_SCHED_NTASKS_THRESHOLD_DEFAULT 2
#define _STARPU_SCHED_EXP_LEN_THRESHOLD_DEFAULT 1000000000.0
static void initialize_eager_prefetching_center_policy(unsigned sched_ctx_id)
{
unsigned ntasks_threshold = _STARPU_SCHED_NTASKS_THRESHOLD_DEFAULT;
double exp_len_threshold = _STARPU_SCHED_EXP_LEN_THRESHOLD_DEFAULT;
[...]
starpu_sched_ctx_create_worker_collection
(sched_ctx_id, STARPU_WORKER_LIST);
/* Create the Scheduling Tree */
struct starpu_sched_tree * t = starpu_sched_tree_create(sched_ctx_id);
/* The Root Component is a Flow-control Fifo Component */
t->root = starpu_sched_component_fifo_create(NULL);
/* The Resource-mapping Component of the strategy is an Eager Component
*/
struct starpu_sched_component *eager_component = starpu_sched_component_eager_create(NULL);
/* Create links between Components : the Eager Component is the child
* of the Root Component */
starpu_sched_component_connect(t->root, eager_component);
/* A task threshold is set for the Flow-control Components which will
* be connected to Worker Components. By doing so, this Modularized
* Scheduler will be able to perform some prefetching on the resources
*/
struct starpu_sched_component_fifo_data fifo_data =
{
.ntasks_threshold = ntasks_threshold,
.exp_len_threshold = exp_len_threshold,
};
unsigned i;
for(i = 0; i < starpu_worker_get_count() + starpu_combined_worker_get_count(); i++)
{
/* Each Worker Component has a Flow-control Fifo Component as
* father */
struct starpu_sched_component * worker_component = starpu_sched_component_worker_new(i);
struct starpu_sched_component * fifo_component = starpu_sched_component_fifo_create(&fifo_data);
starpu_sched_component_connect(fifo_component, worker_component);
/* Each Flow-control Fifo Component associated to a Worker
* Component is linked to the Eager Component as one of its
* children */
starpu_sched_component_connect(eager_component, fifo_component);
}
starpu_sched_tree_update_workers(t);
starpu_sched_ctx_set_policy_data(sched_ctx_id, (void*)t);
}
/* Properly destroy the Scheduling Tree and all its Components */
static void deinitialize_eager_prefetching_center_policy(unsigned sched_ctx_id)
{
struct starpu_sched_tree * tree = (struct starpu_sched_tree*)starpu_sched_ctx_get_policy_data(sched_ctx_id);
starpu_sched_tree_destroy(tree);
starpu_sched_ctx_delete_worker_collection(sched_ctx_id);
}
/* Initializing the starpu_sched_policy struct associated to the Modularized
* Scheduler : only the init_sched and deinit_sched needs to be defined to
* implement a Modularized Scheduler */
struct starpu_sched_policy _starpu_sched_tree_eager_prefetching_policy =
{
.init_sched = initialize_eager_prefetching_center_policy,
.deinit_sched = deinitialize_eager_prefetching_center_policy,
.add_workers = starpu_sched_tree_add_workers,
.remove_workers = starpu_sched_tree_remove_workers,
.push_task = starpu_sched_tree_push_task,
.pop_task = starpu_sched_tree_pop_task,
.pre_exec_hook = starpu_sched_component_worker_pre_exec_hook,
.post_exec_hook = starpu_sched_component_worker_post_exec_hook,
.pop_every_task = NULL,
.policy_name = "tree-eager-prefetching",
.policy_description = "eager with prefetching tree policy"
};
\endcode
Other modular scheduler examples can be seen in src/sched_policies/modular_*.c
For instance, \c modular-heft-prio needs performance models, decides
memory nodes, uses prioritized fifos above and below, and decides the best
implementation.
If unsure on the result of the modular scheduler construction, you can run a
simple application with FxT enabled (see \ref GeneratingTracesWithFxT), and open
the generated file \c trace.html in a web-browser.
\subsection ModularizedSchedulersAndParallelTasks Management of parallel task
At the moment, parallel tasks can be managed in modularized schedulers through
combined workers: instead of connecting a scheduling component to a worker
component, one can connect it to a combined worker component (i.e. a worker
component created with a combined worker id). That component will handle
creating task aliases for parallel execution and push them to the different
workers components.
\subsection WriteASchedulingComponent Writing a Scheduling Component
\subsubsection GenericSchedulingComponent Generic Scheduling Component
Each Scheduling Component is instantiated from a Generic Scheduling Component,
which implements a generic version of the Interface. The generic implementation
of Pull, Can_Pull and Can_Push functions are recursive calls to their parents
(respectively to their children). However, as a Generic Scheduling Component do
not know how much children it will have when it will be instantiated, it does
not implement the Push function.
\subsubsection InstantiationRedefineInterface Instantiation : Redefining the Interface
A Scheduling Component must implement all the functions of the Interface. It is
so necessary to implement a Push function to instantiate a Scheduling Component.
The implemented Push function is the "fingerprint" of a Scheduling Component.
Depending on how functionalities or properties programmers want to give
to the Scheduling Component they are implementing, it is possible to reimplement
all the functions of the Interface. For example, a Flow-control Component
reimplements the Pull and the Can_Push functions of the Interface, allowing
to catch the generic recursive calls of these functions. The Pull function of
a Flow-control Component can, for example, pop a task from the local storage
queue of the Component, and give it to the calling Component which asks for it.
\subsubsection DetailedProgressionAndValidationRules Detailed Progression and Validation Rules
- A Reservoir is a Scheduling Component which redefines a Push and a Pull
function, in order to store tasks into it. A Reservoir delimit Scheduling
Areas in the Scheduling Tree.
- A Pump is the engine source of the Scheduler : it pushes/pulls tasks
to/from a Scheduling Component to an other. Native Pumps of a Scheduling
Tree are located at the root of the Tree (incoming Push calls from StarPU),
and at the leafs of the Tree (Pop calls coming from StarPU Workers).
Pre-implemented Scheduling Components currently shipped with Pumps are
Flow-Control Components and the Resource-Mapping Component Heft, within
their defined Can_Push functions.
- A correct Scheduling Tree requires a Pump per Scheduling Area and per
Execution Flow.
The Tree-Eager-Prefetching Scheduler shown in Section
\ref ImplementAModularizedScheduler follows the previous assumptions :
starpu_push_task
Pump
|
Area 1 |
|
v
-----------------------Fifo_Component-----------------------------
Pump
| ^
Push | | Can_Push
v |
Area 2 Eager_Component
| ^
| |
v |
--------><-------------------><---------
| ^ | ^
Push | | Can_Push Push | | Can_Push
v | v |
-----Fifo_Component-----------------------Fifo_Component----------
| ^ | ^
Pull | | Can_Pull Pull | | Can_Pull
Area 3 v | v |
Pump Pump
Worker_Component Worker_Component
\section GraphScheduling Graph-based Scheduling
For performance reasons, most of the schedulers shipped with StarPU use simple
list-scheduling heuristics, assuming that the application has already set
priorities. This is why they do their scheduling between when tasks become
available for execution and when a worker becomes idle, without looking at the
task graph.
Other heuristics can however look at the task graph. Recording the task graph
is expensive, so it is not available by default, the scheduling heuristic has
to set \c _starpu_graph_record to \c 1 from the initialization function, to make it
available. Then the _starpu_graph* functions can be used.
src/sched_policies/graph_test_policy.c is an example of simple greedy
policy which automatically computes priorities by bottom-up rank.
The idea is that while the application submits tasks, they are only pushed
to a bag of tasks. When the application is finished with submitting tasks,
it calls starpu_do_schedule() (or starpu_task_wait_for_all(), which calls
starpu_do_schedule()), and the starpu_sched_policy::do_schedule method of the
scheduler is called. This method calls \c _starpu_graph_compute_depths() to compute
the bottom-up ranks, and then uses these ranks to set priorities over tasks.
It then has two priority queues, one for CPUs, and one for GPUs, and uses a dumb
heuristic based on the duration of the task over CPUs and GPUs to decide between
the two queues. CPU workers can then pop from the CPU priority queue, and GPU
workers from the GPU priority queue.
\section DebuggingScheduling Debugging Scheduling
All the \ref OnlinePerformanceTools and \ref OfflinePerformanceTools can
be used to get information about how well the execution proceeded, and thus the
overall quality of the execution.
Precise debugging can also be performed by using the
\ref STARPU_TASK_BREAK_ON_PUSH, \ref STARPU_TASK_BREAK_ON_SCHED,
\ref STARPU_TASK_BREAK_ON_POP, and \ref STARPU_TASK_BREAK_ON_EXEC environment variables.
By setting the job_id of a task
in these environment variables, StarPU will raise SIGTRAP when the task is being
scheduled, pushed, or popped by the scheduler. This means that when one notices
that a task is being scheduled in a seemingly odd way, one can just reexecute
the application in a debugger, with some of those variables set, and the
execution will stop exactly at the scheduling points of this task, thus allowing
to inspect the scheduler state, etc.
*/