aserio changed the topic of #ste||ar to: STE||AR: Systems Technology, Emergent Parallelism, and Algorithm Research | stellar.cct.lsu.edu | HPX: A cure for performance impaired parallel applications | github.com/STEllAR-GROUP/hpx | Buildbot: http://rostam.cct.lsu.edu/ | Log: http://irclog.cct.lsu.edu/
EverYoun_ has quit [Read error: Connection reset by peer]
EverYoung has joined #ste||ar
EverYoung has quit [Remote host closed the connection]
EverYoung has joined #ste||ar
EverYoung has quit [Ping timeout: 276 seconds]
diehlpk has joined #ste||ar
EverYoung has joined #ste||ar
EverYoung has quit [Remote host closed the connection]
EverYoung has joined #ste||ar
EverYoung has quit [Ping timeout: 276 seconds]
diehlpk has quit [Ping timeout: 240 seconds]
jaafar has quit [Ping timeout: 256 seconds]
eschnett has quit [Quit: eschnett]
eschnett has joined #ste||ar
hkaiser has quit [Quit: bye]
diehlpk has joined #ste||ar
eschnett has quit [Quit: eschnett]
diehlpk has quit [Remote host closed the connection]
K-ballo has quit [Quit: K-ballo]
hkaiser has joined #ste||ar
anushi has quit [Remote host closed the connection]
nanashi55 has quit [Ping timeout: 240 seconds]
nanashi55 has joined #ste||ar
anushi has joined #ste||ar
hkaiser has quit [Quit: bye]
mcopik has quit [Ping timeout: 260 seconds]
anushi has quit [Ping timeout: 252 seconds]
anushi has joined #ste||ar
anushi has quit [Remote host closed the connection]
nanashi55 has quit [Ping timeout: 255 seconds]
nanashi55 has joined #ste||ar
<github>
[hpx] msimberg force-pushed msimberg-patch-1 from 4634328 to e4f18de: https://git.io/vx1w3
<github>
hpx/msimberg-patch-1 e4f18de Mikael Simberg: Fix bulk_construct with count = 0
<github>
[hpx] msimberg created msimberg-patch-2 (+1 new commit): https://git.io/vx11f
<github>
hpx/msimberg-patch-2 4eff902 Mikael Simberg: Remove leftover code from context_generic_context.hpp...
<github>
[hpx] msimberg opened pull request #3271: Remove leftover code from context_generic_context.hpp (master...msimberg-patch-2) https://git.io/vx11l
mcopik has quit [Ping timeout: 276 seconds]
anushi has quit [Ping timeout: 255 seconds]
anushi has joined #ste||ar
diehlpk_work_ has joined #ste||ar
gentryx_ has joined #ste||ar
gentryx has quit [*.net *.split]
simbergm has quit [*.net *.split]
diehlpk_work has quit [*.net *.split]
simbergm has joined #ste||ar
nikunj_ has joined #ste||ar
anushi has quit [Ping timeout: 240 seconds]
K-ballo has joined #ste||ar
hkaiser has joined #ste||ar
anushi has joined #ste||ar
nikunj_ has quit [Quit: Page closed]
anushi has quit [Ping timeout: 255 seconds]
Smasher has joined #ste||ar
Smasher has quit [Remote host closed the connection]
Smasher has joined #ste||ar
victor_ludorum has joined #ste||ar
nikunj has joined #ste||ar
anushi has joined #ste||ar
anushi has quit [Remote host closed the connection]
<jbjnr_>
hkaiser: you pinged me a few days ago
<hkaiser>
jbjnr_: hey
<hkaiser>
g'morning
<jbjnr_>
greetigs
<hkaiser>
jbjnr_: I would like to write a paper for the rappersville meeting 'Executors should be variadic'
<jbjnr_>
yes. I want to do that too
<hkaiser>
the executor stuf you wanted to write about would fit just nicely as one of the arguments
victor_ludorum has quit [Ping timeout: 260 seconds]
<hkaiser>
I'd like to ask you to join efforts
<jbjnr_>
of coure
<jbjnr_>
course
<hkaiser>
ok, I'll try to get a document number and start putting together some thoughts
<jbjnr_>
I suspect it can still be done with function objects though
<jbjnr_>
just a bit mastier
<K-ballo>
I see that old issue refuses to close
<hkaiser>
not in the general case, only if you know what type of FO you get
<hkaiser>
K-ballo: yah, for a good reason
<jbjnr_>
you can pattern match to any signature though
<hkaiser>
this puts the whole current executor paper in jeopardy ;)
<hkaiser>
jbjnr_: shrug
<jbjnr_>
I have to run away for a mo. off to hospital for more checkups :(
<hkaiser>
this is only one of the arguments in favor of a variadic interface
<hkaiser>
jbjnr_: ok
<K-ballo>
and is there absolutely no way out?
<hkaiser>
out of what?
eschnett has joined #ste||ar
nikunj_ has joined #ste||ar
<K-ballo>
avoiding the variadic interfaces.. there's obvious opposition to that
nikunj has quit [Ping timeout: 260 seconds]
eschnett has quit [Quit: eschnett]
<hkaiser>
K-ballo: the opposition is mostly NVidia as they want to bleed their GPU requirements through the executor API
eschnett has joined #ste||ar
hkaiser has quit [Quit: bye]
daissgr has joined #ste||ar
hkaiser has joined #ste||ar
anushi has joined #ste||ar
nikunj_ has quit [Ping timeout: 260 seconds]
anushi has quit [Ping timeout: 255 seconds]
mbremer has joined #ste||ar
<mbremer>
Hi all, I had a question about where some overheads in HPX are coming from.
<mbremer>
I'm running a stencil-like problem and I have two partitioning schemes. In both cases, the partitionings are load balanced, however, in one I do my best minimize inter-node communication and in the other case I don't really care.
daissgr has quit [Quit: WeeChat 1.9.1]
<mbremer>
As one would expect, the internode-communication version does better (by a factor of 2.4x in some cases)
<mbremer>
I'm wondering what's causing this overhead? I'm using channels to communicate between components. Is there some sort of overhead there that would be present in one case that's not in the other?
<mbremer>
Also are there counters I could measure to try and track down/quantify whatever overhead is in fact causing the slowdown?
jaafar has joined #ste||ar
aserio has joined #ste||ar
anushi has joined #ste||ar
jaafar_ has joined #ste||ar
anushi has quit [Remote host closed the connection]
jaafar has quit [Ping timeout: 240 seconds]
nikunj has joined #ste||ar
anushi has joined #ste||ar
jaafar_ is now known as jaafar
aserio1 has joined #ste||ar
aserio has quit [Ping timeout: 276 seconds]
aserio1 is now known as aserio
<hkaiser>
mbremer: well, networking itself _is_ the overhead, yes?
<mbremer>
Presumably?
<hkaiser>
;)
aserio1 has joined #ste||ar
<mbremer>
So you would think it's something along the lines of the fact that more messages are getting sent is causing the overhead?
<mbremer>
Does that mean that when channels are "sending" messages locally they are able to bypass the parcelport layer?
aserio has quit [Ping timeout: 265 seconds]
aserio1 is now known as aserio
<mbremer>
Also @hkaiser why wouldn't I see the same overheads for my flat MPI implementation?
mbremer has quit [Ping timeout: 260 seconds]
anushi has quit [Read error: Connection reset by peer]
mcopik has joined #ste||ar
EverYoung has joined #ste||ar
anushi has joined #ste||ar
EverYoun_ has joined #ste||ar
EverYoung has quit [Read error: Connection reset by peer]
EverYoun_ has quit [Remote host closed the connection]
EverYoung has joined #ste||ar
aserio has quit [Ping timeout: 240 seconds]
anushi has quit [Ping timeout: 265 seconds]
eschnett has quit [Quit: eschnett]
mbremer has joined #ste||ar
EverYoung has quit [Ping timeout: 265 seconds]
EverYoung has joined #ste||ar
<hkaiser>
mbremer: exactly - local operations don't go through the networking
<mbremer>
Do you suspect that this might have some effect on the scheduler? Something like more hpx threads have to get suspended, and then more stealing happens
<hkaiser>
mbremer: could be, most likely however networking is just generally more expensive
<mbremer>
As in writing to the NIC is more expensive?
<mbremer>
Or how do you mean more expensive?
<hkaiser>
mbremer: networking in hpx requires serialization, which is not needed for the local case
<hkaiser>
and there is also some overhead caused by the impedence mismatch between how HPX works and the underlying networking (MPI) is expecting things to happen
david_pfander has quit [Ping timeout: 256 seconds]
<mbremer>
I see. Kk, that helps. I'll write that up.
<mbremer>
Thanks @hkaiser
<mbremer>
@hkaiser: also, I was thinking for that grant proposal: Have you thought of creating something similar to Charm++'s AMPI?
<hkaiser>
mbremer: heh
<mbremer>
I was thinking that might be a good thing for the proposal. It might let you get a whole set of codes onto an HPX thing, and then you could incrementally modernize the code bases
<hkaiser>
no plans, and not sure if this would be the right thing to do
<hkaiser>
I'd rather allow for a clear partitioning of the available compute resources to carve out a (dynamic) part for the existing MPI code
<mbremer>
I see.
<hkaiser>
this gives you a migration path as well - also using HPX executors to schedule MPI tasks etc.
anushi has joined #ste||ar
<hkaiser>
mbremer: that's why we started with that whole resource partitioner story
mbremer_ has joined #ste||ar
<mbremer_>
@hkaiser: Yeah, it was just a thought.
<hkaiser>
sure
anushi has quit [Remote host closed the connection]
mbremer_ has quit [Ping timeout: 260 seconds]
eschnett has joined #ste||ar
anushi has joined #ste||ar
anushi has quit [Remote host closed the connection]
aserio has joined #ste||ar
eschnett has quit [Ping timeout: 240 seconds]
eschnett has joined #ste||ar
jaafar has quit [Quit: Konversation terminated!]
eschnett has quit [Quit: eschnett]
eschnett has joined #ste||ar
nikunj has quit [Quit: Page closed]
EverYoun_ has joined #ste||ar
<jbjnr_>
hkaiser: yt?
EverYoung has quit [Ping timeout: 240 seconds]
eschnett has quit [Quit: eschnett]
<hkaiser>
here
<hkaiser>
jbjnr_: ... or not - gtg
hkaiser has quit [Quit: bye]
EverYoun_ has quit [Remote host closed the connection]
EverYoung has joined #ste||ar
mbremer has quit [Quit: Page closed]
EverYoung has quit [Remote host closed the connection]
EverYoun_ has joined #ste||ar
hkaiser has joined #ste||ar
eschnett has joined #ste||ar
EverYoung has joined #ste||ar
EverYoun_ has quit [Ping timeout: 255 seconds]
EverYoung has quit [Remote host closed the connection]
aserio has quit [Quit: aserio]
EverYoung has joined #ste||ar
EverYoung has quit [Remote host closed the connection]
EverYoung has joined #ste||ar
EverYoun_ has joined #ste||ar
EverYoung has quit [Ping timeout: 276 seconds]
EverYoun_ has quit [Ping timeout: 276 seconds]
jaafar has joined #ste||ar
EverYoung has joined #ste||ar
EverYoung has quit [Remote host closed the connection]