Q: Since registerFd uses OneShot and threadWait uses registerFd, basic
IO functions use OneShot by default. No changes from GHC 7.8.3. Do
I understand correctly?
That is the idea. That being said adding another variant of registerFd
(which as far as I know has three users) for
Kazu Yamamoto k...@iij.ad.jp writes:
Hi,
Andreas - want me to go ahead and get you some hardware to test Ben's
patch in the mean time? This way we'll at least not leave it hanging
until the last moment...
I will also try this with two 20-core machines connected 10G on
Monday.
I
Kazu Yamamoto k...@iij.ad.jp writes:
Hi,
I measured the performace of GHC head, 7.8.3 and 7.8.3 + Ben's patch
set.
Server: witty 8080 -r -a -s +RTS -Nn *1
Measurement tool: weighttp -n 10 -c 1000 -k -t 19 http://192.168.0.1:8080/
Measurement env: two 20 core (w/o HT) machines
Kazu Yamamoto k...@iij.ad.jp writes:
Hi,
Hi Kazu,
Andreas - want me to go ahead and get you some hardware to test Ben's
patch in the mean time? This way we'll at least not leave it hanging
until the last moment...
I will also try this with two 20-core machines connected 10G on
Monday.
Ben,
Hmm, uh oh. Thanks for testing this. I'll try to reproduce this on my
end. It looks like it shouldn't be so hard as even the single-threaded
performance regresses drastically. Just to confirm, you are using the
latest revision of D347?
I used the following as you suggested:
Ben,
This may be due to lacking INLINEs on definitions added in
GHC.Event.Internal [1]. I'm currently in the middle of reproducing these
results on an EC2 instance to confirm this. So far the results look much
more consistent than my previous attempts at benchmarking on my own
hardware.
If
Kazu Yamamoto k...@iij.ad.jp writes:
Ben,
This may be due to lacking INLINEs on definitions added in
GHC.Event.Internal [1]. I'm currently in the middle of reproducing these
results on an EC2 instance to confirm this. So far the results look much
more consistent than my previous attempts at
I already pushed it. The commit in question is
5dce47eb8415eb31e1c6759b6f6a2ef5bfe32470. Thanks for the benchmarking!
I believe this is in bgamari/ghc (for GHC 7.10?).
I'm using bgamari/packages-base for GHC 7.8 and asking to push the
same commit to this repo.
Actually I compared the latest
Kazu Yamamoto k...@iij.ad.jp writes:
I already pushed it. The commit in question is
5dce47eb8415eb31e1c6759b6f6a2ef5bfe32470. Thanks for the benchmarking!
I believe this is in bgamari/ghc (for GHC 7.10?).
I'm using bgamari/packages-base for GHC 7.8 and asking to push the
same commit to this
Ahh, yes. Sorry, I forgot you were on 7.8. Just pushed a new patch to
the event-rework-squashed branch [1].
I believe that you are trying to merge your patches to GHC 7.8.4?
If not, I will work on the GHC head branch.
--Kazu
___
ghc-devs mailing
Well, Bas was wondering whether this would be possible. At this point
I'm a bit on the fence; on one hand it's not a crucial fix (we have a
workaround in usb) and it may involve changes to exported interfaces
(although not very high visibility). On the other hand, it's a pretty
easy change to
Kazu Yamamoto k...@iij.ad.jp writes:
Well, Bas was wondering whether this would be possible. At this point
I'm a bit on the fence; on one hand it's not a crucial fix (we have a
workaround in usb) and it may involve changes to exported interfaces
(although not very high visibility). On the
Out of curiosity are these numbers from single runs or do you average?
Run three times and took the middle in this time.
What are the uncertainties on these numbers? Even on the Rackspace
machines I was finding very large variances in my benchmarks, largely
due to far outliers. I didn't
Hi Austin,
You need to set the CPU into C0 using /dev/cpu_dma_latency. Here's a
short paper with a program to show the way to do it[1].
This paper is what I'm looking for. Thanks!
--Kazu
___
ghc-devs mailing list
ghc-devs@haskell.org
Hi,
Andreas - want me to go ahead and get you some hardware to test Ben's
patch in the mean time? This way we'll at least not leave it hanging
until the last moment...
I will also try this with two 20-core machines connected 10G on
Monday.
I measured the performace of GHC head, 7.8.3 and
Hi Ben, Austin,
Is there any chance of Ben's event manager patch landing in GHC-7.8.4?
Bas
On 13 October 2014 21:05, Ben Gamari bgamari.f...@gmail.com wrote:
Ben Gamari bgamari.f...@gmail.com writes:
Andreas Voellmy andreas.voel...@gmail.com writes:
On Sat, Oct 11, 2014 at 12:17 PM, Ben
I haven't had a chance to dig into Ben's patch yet, but I expect it will
accepted soon - I don't think the change will affect performance.
Austin, would it be possible to get a relatively minor change to the event
manager into 7.8.4? It may change a semi-public API under GHC.Event, but
will not
The catch with such a change is that there is no macro to determine
whether we're using 7.8.3 or 7.8.4, so it's harder for users to figure
things out (they have to use `MIN_VERSION_base` from Cabal). But maybe
that doesn'tm atter too much. So, yes, I think it's doable, but that's
a sticky bit.
Austin,
Andreas - want me to go ahead and get you some hardware to test Ben's
patch in the mean time? This way we'll at least not leave it hanging
until the last moment...
I will also try this with two 20-core machines connected 10G on
Monday.
--Kazu
Austin Seipp aus...@well-typed.com writes:
The catch with such a change is that there is no macro to determine
whether we're using 7.8.3 or 7.8.4, so it's harder for users to figure
things out (they have to use `MIN_VERSION_base` from Cabal). But maybe
that doesn'tm atter too much. So, yes, I
Hi,
On 17 October 2014 16:27, Austin Seipp aus...@well-typed.com wrote:
The catch with such a change is that there is no macro to determine
whether we're using 7.8.3 or 7.8.4, so it's harder for users to figure
things out (they have to use `MIN_VERSION_base` from Cabal). But maybe
that
Austin Seipp aus...@well-typed.com writes:
The catch with such a change is that there is no macro to determine
whether we're using 7.8.3 or 7.8.4, so it's harder for users to figure
things out (they have to use `MIN_VERSION_base` from Cabal). But maybe
that doesn'tm atter too much. So, yes, I
On 2014-10-13 at 23:33:13 +0200, Austin Seipp wrote:
[...]
Also, if any other developers (like Andreas, Johan, Bryan, etc) in
this space want a big machine to test it on, I can probably equip you
with one (or several). Since Rackspace is so gracious to us, we can
immediately allocate
This is awesome. I'd like to try to recreate some of the evaluations for
the multicore IO manager paper on that 40 core system at backspace. How can
I get access to this? I'll jump on IRC - maybe it is easier to chat in
realtime.
On Mon, Oct 13, 2014 at 5:33 PM, Austin Seipp aus...@well-typed.com
Hey Andreas,
The basic rundown is that if we equip you with an account, you can
just do it yourself. Although we'd like to restrict access a bit more;
I'll figure something out.
Yeah, if you hop on IRC, we can chat quickly about it and work
something out in the mean time.
On Tue, Oct 14, 2014
Andreas Voellmy andreas.voel...@gmail.com writes:
This is awesome. I'd like to try to recreate some of the evaluations for
the multicore IO manager paper on that 40 core system at backspace. How can
I get access to this? I'll jump on IRC - maybe it is easier to chat in
realtime.
Do you
Yes, I'll try to describe it and script it so that others can understand
the benchmarks and run it easily as well.
On Tue, Oct 14, 2014 at 1:23 PM, Ben Gamari bgamari.f...@gmail.com wrote:
Andreas Voellmy andreas.voel...@gmail.com writes:
This is awesome. I'd like to try to recreate some of
Ben Gamari bgamari.f...@gmail.com writes:
Andreas Voellmy andreas.voel...@gmail.com writes:
On Sat, Oct 11, 2014 at 12:17 PM, Ben Gamari bgamari.f...@gmail.com wrote:
Ah... so this is not useful to you. I guess we could add `loop` to
GHC.Event's export list. On the other hand, I like your
For the record, I talked to Ben earlier on IRC, and I can provide him
with a machine to do intense benchmarks of the new I/O manager.
Also, if any other developers (like Andreas, Johan, Bryan, etc) in
this space want a big machine to test it on, I can probably equip you
with one (or several).
On Sat, Oct 11, 2014 at 1:07 AM, Ben Gamari bgamari.f...@gmail.com wrote:
Thanks for your quick reply!
Andreas Voellmy andreas.voel...@gmail.com writes:
On Sat, Oct 11, 2014 at 12:17 AM, Ben Gamari bgamari.f...@gmail.com
wrote:
I'm a bit perplexed as to why the change was made in the
Another way to fix usb would be to re-register the callback after a
previously registered callback is fired. Of course it is cheaper not to
have to re-register, but re-registration in the latest IO manager should be
fairly cheap, so this may not be a performance problem for usb. Would this
work
Andreas Voellmy andreas.voel...@gmail.com writes:
Another way to fix usb would be to re-register the callback after a
previously registered callback is fired. Of course it is cheaper not to
have to re-register, but re-registration in the latest IO manager should be
fairly cheap, so this may
Andreas Voellmy andreas.voel...@gmail.com writes:
On Sat, Oct 11, 2014 at 1:07 AM, Ben Gamari bgamari.f...@gmail.com wrote:
Thanks for your quick reply!
What I'm wondering is what the extra complexity bought us. It seems like
the same thing could have been achieved with less breakage by
On Sat, Oct 11, 2014 at 12:17 PM, Ben Gamari bgamari.f...@gmail.com wrote:
Andreas Voellmy andreas.voel...@gmail.com writes:
On Sat, Oct 11, 2014 at 1:07 AM, Ben Gamari bgamari.f...@gmail.com
wrote:
Thanks for your quick reply!
What I'm wondering is what the extra complexity bought
Andreas Voellmy andreas.voel...@gmail.com writes:
On Sat, Oct 11, 2014 at 12:17 PM, Ben Gamari bgamari.f...@gmail.com wrote:
Yes, but it will be invoked by GHC.Thread and any other callers of it will
simply block indefinitely waiting for the thread that is running loop to
give it up - which
On Sat, Oct 11, 2014 at 12:17 AM, Ben Gamari bgamari.f...@gmail.com wrote:
In ba2555ef and a6f52b19 one-shot semantics were added to event manager
in `base`. If my understanding of this code is correct, in this mode the
event manager will use only notify the user of the first event on a
Thanks for your quick reply!
Andreas Voellmy andreas.voel...@gmail.com writes:
On Sat, Oct 11, 2014 at 12:17 AM, Ben Gamari bgamari.f...@gmail.com wrote:
I'm a bit perplexed as to why the change was made in the way that it
was. Making one-shot a event-manager-wide attribute seems to add a
37 matches
Mail list logo