hkaiser changed the topic of #ste||ar to: STE||AR: Systems Technology, Emergent Parallelism, and Algorithm Research | stellar-group.org | HPX: A cure for performance impaired parallel applications | github.com/STEllAR-GROUP/hpx | This channel is logged: irclog.cct.lsu.edu
K-ballo has quit [Quit: K-ballo]
hkaiser has quit [Quit: Bye!]
chaitanya710 has joined #ste||ar
chaitanya710 has quit [Ping timeout: 256 seconds]
chaitanya710 has joined #ste||ar
Ahmedehabb has joined #ste||ar
<Ahmedehabb> hello , i have made a tentative proposal for the Implement a Faster Associative Container for GIDs project.
<Ahmedehabb> should i send it here or privately ?
Ahmedehabb has quit [Quit: Client closed]
Ahmedehabb has joined #ste||ar
K-ballo has joined #ste||ar
Ahmedehabb has quit [Quit: Client closed]
hkaiser has joined #ste||ar
<srinivasyadav227> hkaiser: is `--hpx:numa-sensitive`(https://hpx-docs.stellar-group.org/latest/html/manual/hpx_runtime_and_resources.html#hpx-runtime-and-resources) the one you told about work stealing related to NUMA in the meeting ?
<hkaiser> srinivasyadav227: not sure anymore ;-)
<hkaiser> this options is on by default anyways, so will not have any measurable effect
<srinivasyadav227> i tried using this, for the mandelbrot, but it didnt work
<srinivasyadav227> yeah, there was no effect
<hkaiser> nod
<hkaiser> I created a PR that might improve things, could you try that?
<srinivasyadav227> i saw your branch, https://github.com/STEllAR-GROUP/hpx/tree/numa_stealing
<hkaiser> #5825
<hkaiser> yes
<hkaiser> it might have no effect, please try it out
<srinivasyadav227> yea, i was about to link that none , i will try with this
<hkaiser> thanks
<srinivasyadav227> i compared hpx::compute::vector with std::vector as well but they dont see to have any improvements,
<hkaiser> ok, I didn't expect for compute::vector to make a big difference, it's the allocator/executor that should change things
<srinivasyadav227> yes, i mean, hpx with allocator's and executor's VS plain std::vector without allocator and executors
<hkaiser> ok
<srinivasyadav227> hkaiser: this is the scaling for mandelbrot on ookami (48 cores), only parallelization
<srinivasyadav227> it has 4 numa domains, each containing 12 cores
<srinivasyadav227> till 1 numa domain (12 cores), 75% is being parallelized; till 2 numa domains (24 cores) its around 60-65, after that it drops to 50% parallelization
<hkaiser> yah, cross numa-traffic
<hkaiser> so something is off with the allocator/executor - those should take care of data locality
<hkaiser> srinivasyadav227: gtg now, sorry
<srinivasyadav227> np :), i will try to profile the application till then
hkaiser has quit [Quit: Bye!]
chaitanya710 has quit [Quit: Client closed]
diehlpk_work_ has quit [Ping timeout: 240 seconds]
Ahmedehabb has joined #ste||ar
Ahmedehabb has quit [Quit: Client closed]
Ahmedehabb has joined #ste||ar
diehlpk_work_ has joined #ste||ar
K-ballo has quit [Quit: K-ballo]
K-ballo has joined #ste||ar
Ahmedehabb has quit [Quit: Client closed]