The java process I'm launching is Elasticsearch and it has numerous threads that are created.
I check that using ps HuH p <pid> | wc -l.
this is how I get the pid of elasticsearch:
ES=`jps | egrep 'Elasticsearch' | awk '{print $1}'`
I pinning all the threads from the given pid to a set of cores using the following python script.
#!/usr/bin/env python
import os
import argparse
def task_set(pid, core):
print "pinning all threads of pid: ", pid, " to cores: ", core
os.system('taskset -apc '+str(core)+' ' +str(pid)+' >/dev/null')
def main(args):
experiments = ["1B", "2B", "1B3S", "2B2S", "1S", "2S", "3S", "4S"]
which = args.id[0]
idx = experiments.index(which)
PC = ["0", "0,1", "0,2,3,4", "0,1,2,3", "2", "2,3","2,3,4", "2,3,4,5"]
task_set(args.pid[0], PC[idx])
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--pid', nargs=1, help='appid')
parser.add_argument('--id', nargs=1, help='core')
args = parser.parse_args()
main(args)
However, when I look at top -H -p <pid> and monitor the core allocation. It does not obey most of the time.
Is there something I'm missing here? Does JVM have to do something here?
Is there anyother way I can pin the threads to the cores?