Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
add python bindings for plugin host load
authorAdrien Gougeon <adrien.gougeon@ens-rennes.fr>
Thu, 8 Jun 2023 14:28:08 +0000 (16:28 +0200)
committerAdrien Gougeon <adrien.gougeon@ens-rennes.fr>
Thu, 8 Jun 2023 14:28:08 +0000 (16:28 +0200)
MANIFEST.in
examples/python/CMakeLists.txt
examples/python/plugin-host-load/plugin-host-load.py [new file with mode: 0644]
examples/python/plugin-host-load/plugin-host-load.tesh [new file with mode: 0644]
src/bindings/python/simgrid_python.cpp

index 5fac1c7..453c595 100644 (file)
@@ -482,6 +482,8 @@ include examples/python/platform-failures/platform-failures.py
 include examples/python/platform-failures/platform-failures.tesh
 include examples/python/platform-profile/platform-profile.py
 include examples/python/platform-profile/platform-profile.tesh
+include examples/python/plugin-host-load/plugin-host-load.py
+include examples/python/plugin-host-load/plugin-host-load.tesh
 include examples/python/synchro-barrier/synchro-barrier.py
 include examples/python/synchro-barrier/synchro-barrier.tesh
 include examples/python/synchro-mutex/synchro-mutex.py
index f4afe5b..555cfc1 100644 (file)
@@ -5,6 +5,7 @@ foreach(example actor-create actor-daemon actor-join actor-kill actor-migrate ac
         exec-async exec-basic exec-dvfs exec-remote exec-ptask
         task-io task-simple task-switch-host task-variable-load
         platform-comm-serialize platform-profile platform-failures
+        plugin-host-load
         network-nonlinear clusters-multicpu io-degradation exec-cpu-nonlinear
         synchro-barrier synchro-mutex synchro-semaphore)
   set(tesh_files    ${tesh_files}   ${CMAKE_CURRENT_SOURCE_DIR}/${example}/${example}.tesh)
diff --git a/examples/python/plugin-host-load/plugin-host-load.py b/examples/python/plugin-host-load/plugin-host-load.py
new file mode 100644 (file)
index 0000000..3b6156c
--- /dev/null
@@ -0,0 +1,86 @@
+# Copyright (c) 2006-2023. The SimGrid Team. All rights reserved.
+#
+# This program is free software; you can redistribute it and/or modify it
+# under the terms of the license (GNU LGPL) which comes with this package.
+
+
+from argparse import ArgumentParser
+import sys
+from simgrid import Engine, Host, this_actor, Actor
+
+def parse():
+    parser = ArgumentParser()
+    parser.add_argument(
+        '--platform',
+        type=str,
+        required=True,
+        help='path to the platform description'
+    )
+    return parser.parse_args()
+
+def execute_load_test():
+  host = Host.by_name('MyHost1')
+  this_actor.info(f'Initial peak speed: {host.speed:.0E} flop/s; number of flops computed so far: {host.computed_flops:.0E} (should be 0) and current average load: {host.avg_load} (should be 0)')
+
+  start = Engine.clock
+  this_actor.info('Sleep for 10 seconds')
+  this_actor.sleep_for(10)
+
+  speed = host.speed
+  this_actor.info(f'Done sleeping {Engine.clock - start}s; peak speed: {host.speed:.0E} flop/s; number of flops computed so far: {host.computed_flops:.0E} (nothing should have changed)')
+
+  # Run an activity
+  start = e.clock
+  this_actor.info(f'Run an activity of {200E6:.0E} flops at current speed of {host.speed:.0E} flop/s')
+  this_actor.execute(200E6)
+
+  this_actor.info(f'Done working on my activity; this took {Engine.clock - start}s; current peak speed: {host.speed:.0E} flop/s (when I started the computation, \
+the speed was set to {speed:.0E} flop/s); number of flops computed so \
+far: {host.computed_flops:.0E}, average load as reported by the HostLoad plugin: {host.avg_load:.5f} (should be {200E6 / (10.5 * speed * host.core_count + (Engine.clock - start - 0.5) * host.speed * host.core_count):.5f})')
+
+  # ========= Change power peak =========
+  pstate = 1
+  host.pstate = pstate
+  this_actor.info(f'========= Requesting pstate {pstate} (speed should be of {host.pstate_speed(pstate):.0E} flop/s and is of {host.speed:.0E} flop/s, average load is {host.avg_load:.5f})')
+
+  # Run a second activity
+  start = Engine.clock
+  this_actor.info(f'Run an activity of {100E6:.0E} flops')
+  this_actor.execute(100E6)
+  this_actor.info(f'Done working on my activity; this took {Engine.clock - start}s; current peak speed: {host.speed:.0E} flop/s; number of flops computed so far: {host.computed_flops:.0E}')
+  Engine
+  start = Engine.clock
+  this_actor.info("========= Requesting a reset of the computation and load counters")
+  host.reset_load()
+  this_actor.info(f'After reset: {host.computed_flops:.0E} flops computed; load is {host.avg_load}')
+  this_actor.info('Sleep for 4 seconds')
+  this_actor.sleep_for(4)
+  this_actor.info(f'Done sleeping {Engine.clock - start}s; peak speed: {host.speed:.0E} flop/s; number of flops computed so far: {host.computed_flops:.0E}')
+
+  # =========== Turn the other host off ==========
+  host2 = Host.by_name('MyHost2')
+  this_actor.info(f'Turning MyHost2 off, and sleeping another 10 seconds. MyHost2 computed {host2.computed_flops:.0E} flops so far and has an average load of {host2.avg_load}')
+  host2.turn_off()
+  start = Engine.clock
+  this_actor.sleep_for(10)
+  this_actor.info(f'Done sleeping {Engine.clock - start}s; peak speed: {host.speed:.0E} flop/s; number of flops computed so far: {host.computed_flops:.0E}')
+
+def change_speed():
+  host = Host.by_name('MyHost1')
+  this_actor.sleep_for(10.5)
+  this_actor.info("I slept until now, but now I'll change the speed of this host while the other actor is still computing! This should slow the computation down.")
+  host.pstate = 2
+
+if __name__ == '__main__':
+  args = parse()
+  Host.sg_host_load_plugin_init()
+  e = Engine(sys.argv)
+  e.load_platform(args.platform)
+
+  Actor.create('load_test', e.host_by_name('MyHost1'), execute_load_test)
+  Actor.create('change_speed', e.host_by_name('MyHost1'), change_speed)
+
+  e.run()
+
+  this_actor.info(f'Total simulation time: {Engine.clock}')
+
diff --git a/examples/python/plugin-host-load/plugin-host-load.tesh b/examples/python/plugin-host-load/plugin-host-load.tesh
new file mode 100644 (file)
index 0000000..ad14d19
--- /dev/null
@@ -0,0 +1,21 @@
+#!/usr/bin/env tesh
+
+p This tests the Host Load plugin (that allows the user to get the current load of a host and the computed flops)
+
+$ ${pythoncmd:=python3} ${PYTHON_TOOL_OPTIONS:=} ${srcdir:=.}/plugin-host-load.py --platform ${platfdir}/energy_platform.xml
+> [MyHost1:load_test:(1) 0.000000] [python/INFO] Initial peak speed: 1E+08 flop/s; number of flops computed so far: 0E+00 (should be 0) and current average load: 0.0 (should be 0)
+> [MyHost1:load_test:(1) 0.000000] [python/INFO] Sleep for 10 seconds
+> [MyHost1:load_test:(1) 10.000000] [python/INFO] Done sleeping 10.0s; peak speed: 1E+08 flop/s; number of flops computed so far: 0E+00 (nothing should have changed)
+> [MyHost1:load_test:(1) 10.000000] [python/INFO] Run an activity of 2E+08 flops at current speed of 1E+08 flop/s
+> [MyHost1:change_speed:(2) 10.500000] [python/INFO] I slept until now, but now I'll change the speed of this host while the other actor is still computing! This should slow the computation down.
+> [MyHost1:load_test:(1) 18.000000] [python/INFO] Done working on my activity; this took 8.0s; current peak speed: 2E+07 flop/s (when I started the computation, the speed was set to 1E+08 flop/s); number of flops computed so far: 2E+08, average load as reported by the HostLoad plugin: 0.04167 (should be 0.04167)
+> [MyHost1:load_test:(1) 18.000000] [python/INFO] ========= Requesting pstate 1 (speed should be of 5E+07 flop/s and is of 5E+07 flop/s, average load is 0.04167)
+> [MyHost1:load_test:(1) 18.000000] [python/INFO] Run an activity of 1E+08 flops
+> [MyHost1:load_test:(1) 20.000000] [python/INFO] Done working on my activity; this took 2.0s; current peak speed: 5E+07 flop/s; number of flops computed so far: 3E+08
+> [MyHost1:load_test:(1) 20.000000] [python/INFO] ========= Requesting a reset of the computation and load counters
+> [MyHost1:load_test:(1) 20.000000] [python/INFO] After reset: 0E+00 flops computed; load is 0.0
+> [MyHost1:load_test:(1) 20.000000] [python/INFO] Sleep for 4 seconds
+> [MyHost1:load_test:(1) 24.000000] [python/INFO] Done sleeping 4.0s; peak speed: 5E+07 flop/s; number of flops computed so far: 0E+00
+> [MyHost1:load_test:(1) 24.000000] [python/INFO] Turning MyHost2 off, and sleeping another 10 seconds. MyHost2 computed 0E+00 flops so far and has an average load of 0.0
+> [MyHost1:load_test:(1) 34.000000] [python/INFO] Done sleeping 10.0s; peak speed: 5E+07 flop/s; number of flops computed so far: 0E+00
+> [34.000000] [python/INFO] Total simulation time: 34.0
index 5f0f335..f9fcf3c 100644 (file)
@@ -11,6 +11,7 @@
 #include "simgrid/kernel/ProfileBuilder.hpp"
 #include "simgrid/kernel/routing/NetPoint.hpp"
 #include <simgrid/Exception.hpp>
+#include <simgrid/plugins/load.h>
 #include <simgrid/plugins/task.hpp>
 #include <simgrid/s4u/Actor.hpp>
 #include <simgrid/s4u/Barrier.hpp>
@@ -441,6 +442,8 @@ PYBIND11_MODULE(simgrid, m)
            py::overload_cast<const std::string&, const std::string&, const std::string&>(&Host::create_disk),
            py::call_guard<py::gil_scoped_release>(), "Create a disk")
       .def("seal", &Host::seal, py::call_guard<py::gil_scoped_release>(), "Seal this host")
+      .def("turn_off", &Host::turn_off, py::call_guard<py::gil_scoped_release>(), "Turn off this host")
+      .def("turn_on", &Host::turn_on, py::call_guard<py::gil_scoped_release>(), "Turn on this host")
       .def_property("pstate", &Host::get_pstate,
                     py::cpp_function(&Host::set_pstate, py::call_guard<py::gil_scoped_release>()),
                     "The current pstate (read/write property).")
@@ -476,7 +479,24 @@ PYBIND11_MODULE(simgrid, m)
           "")
       .def(
           "__repr__", [](const Host* h) { return "Host(" + h->get_name() + ")"; },
-          "Textual representation of the Host");
+          "Textual representation of the Host.")
+      .def_static(
+          "sg_host_load_plugin_init", []() { sg_host_load_plugin_init(); }, py::call_guard<py::gil_scoped_release>(),
+          "Initialize host load plugin.")
+      .def(
+          "reset_load", [](const Host* h) { sg_host_load_reset(h); }, py::call_guard<py::gil_scoped_release>(),
+          "Reset counters of the host load plugin for this host.")
+      .def_property_readonly(
+          "current_load", [](const Host* h) { return sg_host_get_current_load(h); }, "Current load of the host.")
+      .def_property_readonly(
+          "avg_load", [](const Host* h) { return sg_host_get_avg_load(h); }, "Average load of the host.")
+      .def_property_readonly(
+          "idle_time", [](const Host* h) { return sg_host_get_idle_time(h); }, "Idle time of the host")
+      .def_property_readonly(
+          "total_idle_time", [](const Host* h) { return sg_host_get_total_idle_time(h); },
+          "Total idle time of the host.")
+      .def_property_readonly(
+          "computed_flops", [](const Host* h) { return sg_host_get_computed_flops(h); }, "Computed flops of the host.");
 
   py::enum_<simgrid::s4u::Host::SharingPolicy>(host, "SharingPolicy")
       .value("NONLINEAR", simgrid::s4u::Host::SharingPolicy::NONLINEAR)