--- Begin Message ---
|
Subject: |
[PATCH] store: Use buffered I/O for all protocol writes |
|
Date: |
Fri, 5 Jun 2020 11:11:51 +0200 |
|
User-agent: |
Mutt/1.10.1 (2018-07-13) |
Hi,
I did some digging to investigate why SSH-based guix operations are currently
so slow. One of the reasons seems to be that some operations write single words
to the socket and due to NOWAIT these result in a lot of small packets.
The attached patch modifies store.scm to buffer all writes, so that won’t happen
any more. I’m seeing about ~40% speedup (6.194s vs 10.075s for the best out of
five runs) for `GUIX_DAEMON_SOCKET=ssh://localhost guix environment guix --
true`, but it also seems to have a negative impact on UNIX domain socket
communication by ~10% (1.561s vs 1.385s). For the notorious r-learnr package
it’s still ~20% better (ssh, 16.620/21.418) and only ~3% worse (unix,
11.489/11.199). Keep in mind localhost has a much larger MTU and much lower
latency than usual networks though. It might benefit from having an even bigger
write buffer (64k vs. currently 8k).
Cheers,
Lars
0001-store-Use-buffered-I-O-for-all-protocol-writes.patch
Description: Text Data
signature.asc
Description: PGP signature
--- End Message ---
--- Begin Message ---
|
Subject: |
Re: [bug#41720] [PATCH] store: Use buffered I/O for all protocol writes |
|
Date: |
Tue, 09 Jun 2020 09:49:49 +0200 |
|
User-agent: |
Gnus/5.13 (Gnus v5.13) Emacs/26.3 (gnu/linux) |
Hi!
Lars-Dominik Braun <ldb@leibniz-psychology.org> skribis:
>> Nice, though the slowdown on Unix-domain sockets is annoying given that
>> it’s the primary transport. Are the 10% significant and stable over
>> multiple runs?
> I don’t want to blind your eyes with meaningless statistics here, but let’s do
> this properly then. So for `guix environment guix -- true` with 10 trials
> these
> are the results:
>
> ---snip---
> $ ministat -A -c 99 before_ssh after_ssh
> x before_ssh
> + after_ssh
> N Min Max Median Avg Stddev
> x 10 7.63 10.635 10.1585 9.9382 0.85517585
> + 10 5.986 6.424 6.1755 6.1917 0.15398416
> Difference at 99.0% confidence
> -3.7465 +/- 0.790815
> -37.698% +/- 5.0788%
> (Student's t, pooled s = 0.614425)
>
> $ ministat -A -c 99 before_unix after_unix
> x before_unix
> + after_unix
> N Min Max Median Avg Stddev
> x 10 1.528 1.624 1.5445 1.5636 0.035842867
> + 10 1.576 1.701 1.6145 1.6229 0.037869513
> Difference at 99.0% confidence
> 0.0593 +/- 0.0474548
> 3.79253% +/- 3.08989%
> (Student's t, pooled s = 0.0368701)
> ---snap---
>
> And the same for `guix environment --ad-hoc r-learnr`:
>
> ---snip---
> $ ministat -A -c 99 before_ssh after_ssh
> x before_ssh
> + after_ssh
> N Min Max Median Avg Stddev
> x 10 20.104 21.629 21.082 21.0115 0.45142472
> + 10 16.082 16.713 16.292 16.3416 0.25018224
> Difference at 99.0% confidence
> -4.6699 +/- 0.469719
> -22.2254% +/- 1.86734%
> (Student's t, pooled s = 0.364949)
>
> $ ministat -A -c 99 before_unix after_unix
> x before_unix
> + after_unix
> N Min Max Median Avg Stddev
> x 10 11.322 11.529 11.401 11.4127 0.062467858
> + 10 11.249 11.469 11.3515 11.368 0.078689262
> No difference proven at 99.0% confidence
> ---snap---
Awesome, thanks a lot. Looks like the difference on Unix-domain sockets
is not significant.
Applied!
What package provides ‘ministat’?
Thanks,
Ludo’.
--- End Message ---