* Further `g' tweaks
@ 2010-09-22 22:06 Lars Magne Ingebrigtsen
2010-09-22 22:27 ` Lars Magne Ingebrigtsen
2010-09-23 8:12 ` David Engster
0 siblings, 2 replies; 11+ messages in thread
From: Lars Magne Ingebrigtsen @ 2010-09-22 22:06 UTC (permalink / raw)
To: ding
I'm oscillating a bit back and forth between different `g' tweaks, but
Gnus had gone back to statting thousands of files again, so I did some
burrowing down and tried to find the reason.
It's back to 200 stats again for me, down from 5K stats earlier this
evening.
If I broke something again, let me know. :-)
--
(domestic pets only, the antidote for overdose, milk.)
larsi@gnus.org * Lars Magne Ingebrigtsen
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Further `g' tweaks
2010-09-22 22:06 Further `g' tweaks Lars Magne Ingebrigtsen
@ 2010-09-22 22:27 ` Lars Magne Ingebrigtsen
2010-09-22 23:18 ` Dan Christensen
2010-09-23 8:12 ` David Engster
1 sibling, 1 reply; 11+ messages in thread
From: Lars Magne Ingebrigtsen @ 2010-09-22 22:27 UTC (permalink / raw)
To: ding
Lars Magne Ingebrigtsen <larsi@gnus.org> writes:
> It's back to 200 stats again for me, down from 5K stats earlier this
> evening.
A `g' now takes me 0.8s. That's with a pop server, two nntp servers and
three imap servers, plus the usual nnml, nnfolder and nnvirtual stuff.
--
(domestic pets only, the antidote for overdose, milk.)
larsi@gnus.org * Lars Magne Ingebrigtsen
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Further `g' tweaks
2010-09-22 22:27 ` Lars Magne Ingebrigtsen
@ 2010-09-22 23:18 ` Dan Christensen
2010-09-23 0:36 ` Dan Christensen
0 siblings, 1 reply; 11+ messages in thread
From: Dan Christensen @ 2010-09-22 23:18 UTC (permalink / raw)
To: ding
Lars Magne Ingebrigtsen <larsi@gnus.org> writes:
> Lars Magne Ingebrigtsen <larsi@gnus.org> writes:
>
>> It's back to 200 stats again for me, down from 5K stats earlier this
>> evening.
>
> A `g' now takes me 0.8s. That's with a pop server, two nntp servers and
> three imap servers, plus the usual nnml, nnfolder and nnvirtual stuff.
Starting up gnus takes 0.56s for me, using (gnus 2 t). This involves
a local dovecot server, an nnfolder server and one nnvirtual group.
Hitting `g' then takes 3.5s. This is down from 10.0s with an old
version Gnus. This opens three nntp servers in addition.
A second `g' takes 0.96s, with the servers already open.
I'm pretty happy with that, especially the initial startup.
Dan
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Further `g' tweaks
2010-09-22 23:18 ` Dan Christensen
@ 2010-09-23 0:36 ` Dan Christensen
2010-09-23 15:55 ` Lars Magne Ingebrigtsen
0 siblings, 1 reply; 11+ messages in thread
From: Dan Christensen @ 2010-09-23 0:36 UTC (permalink / raw)
To: ding
Whoops, the numbers below were with an *ancient* version (at least 2 or
3 hours old :-). Updates below, as well as a trace:
Dan Christensen <jdc@uwo.ca> writes:
> Starting up gnus takes 0.56s for me, using (gnus 2 t). This involves
> a local dovecot server, an nnfolder server and one nnvirtual group.
Maybe that was just a lucky run, because now it's more like 0.7 to 0.9s.
Or maybe a recent change slowed it down a touch?
Switching to a shell connection to my local dovecot server brings it
down to 0.45s.
> Hitting `g' then takes 3.5s. This is down from 10.0s with an old
> version Gnus. This opens three nntp servers in addition.
This is more like 3.2 or 3.3s now.
> A second `g' takes 0.96s, with the servers already open.
About 0.8s now.
*gnus trace* buffer:
(gnus 2 t):
20:17:54 (nnimap "rocky" (nnimap-address "localhost") (nnimap-server-port 143) (nnimap-stream network) (nnimap-nov-is-evil t))
20:17:54 (nnfolder "")
20:17:54 (nnvirtual "nnimap\\+rocky:\\(INBOX\\|Sent\\)")
3 g:
20:18:22 (nntp "news.gmane.org")
20:18:22 (nntp "news.astraweb.com")
20:18:23 (nntp "news.gwene.org")
20:18:24 (nntp "news.gmane.org")
20:18:24 (nntp "news.astraweb.com")
20:18:25 (nntp "news.gwene.org")
3 g again:
20:18:31 (nntp "news.gmane.org")
20:18:31 (nntp "news.astraweb.com")
20:18:31 (nntp "news.gwene.org")
20:18:31 (nntp "news.gmane.org")
20:18:31 (nntp "news.astraweb.com")
20:18:31 (nntp "news.gwene.org")
Dan
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Further `g' tweaks
2010-09-23 0:36 ` Dan Christensen
@ 2010-09-23 15:55 ` Lars Magne Ingebrigtsen
2010-09-23 18:31 ` Dan Christensen
0 siblings, 1 reply; 11+ messages in thread
From: Lars Magne Ingebrigtsen @ 2010-09-23 15:55 UTC (permalink / raw)
To: ding
Dan Christensen <jdc@uwo.ca> writes:
>> Hitting `g' then takes 3.5s. This is down from 10.0s with an old
>> version Gnus. This opens three nntp servers in addition.
>
> This is more like 3.2 or 3.3s now.
Right. So opening servers dominates the startup time pretty
dramatically.
> 3 g:
>
> 20:18:22 (nntp "news.gmane.org")
> 20:18:22 (nntp "news.astraweb.com")
> 20:18:23 (nntp "news.gwene.org")
I wonder whether it's possible to somehow multithread the server opening
thing -- opening them in parallel somehow? I don't quite see how,
though.
--
(domestic pets only, the antidote for overdose, milk.)
larsi@gnus.org * Lars Magne Ingebrigtsen
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Further `g' tweaks
2010-09-23 15:55 ` Lars Magne Ingebrigtsen
@ 2010-09-23 18:31 ` Dan Christensen
2010-09-23 18:39 ` Lars Magne Ingebrigtsen
0 siblings, 1 reply; 11+ messages in thread
From: Dan Christensen @ 2010-09-23 18:31 UTC (permalink / raw)
To: ding
Lars Magne Ingebrigtsen <larsi@gnus.org> writes:
> Dan Christensen <jdc@uwo.ca> writes:
>
>>> Hitting `g' then takes 3.5s. This is down from 10.0s with an old
>>> version Gnus. This opens three nntp servers in addition.
>>
>> This is more like 3.2 or 3.3s now.
>
> Right. So opening servers dominates the startup time pretty
> dramatically.
Yes. I wonder why opening a server takes so much more time than
refreshing the information from a server.
>> 3 g:
>>
>> 20:18:22 (nntp "news.gmane.org")
>> 20:18:22 (nntp "news.astraweb.com")
>> 20:18:23 (nntp "news.gwene.org")
>
> I wonder whether it's possible to somehow multithread the server opening
> thing -- opening them in parallel somehow? I don't quite see how,
> though.
That would be cool. I thought your recent magic did that, but maybe is
was only for `g'?
By the way, does it make sense that each server is listed twice in
the trace output when I hit `3 g' once?
Dan
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Further `g' tweaks
2010-09-23 18:31 ` Dan Christensen
@ 2010-09-23 18:39 ` Lars Magne Ingebrigtsen
0 siblings, 0 replies; 11+ messages in thread
From: Lars Magne Ingebrigtsen @ 2010-09-23 18:39 UTC (permalink / raw)
To: ding
Dan Christensen <jdc@uwo.ca> writes:
>> I wonder whether it's possible to somehow multithread the server opening
>> thing -- opening them in parallel somehow? I don't quite see how,
>> though.
>
> That would be cool. I thought your recent magic did that, but maybe is
> was only for `g'?
Yup. Opening the servers is totally synchronous.
> By the way, does it make sense that each server is listed twice in
> the trace output when I hit `3 g' once?
I think that's OK -- it loops through all the servers twice. The
-server-open call by itself takes almost no time (well, unless it
actually has to make a connection). It just swaps around some
variables.
--
(domestic pets only, the antidote for overdose, milk.)
larsi@gnus.org * Lars Magne Ingebrigtsen
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Further `g' tweaks
2010-09-22 22:06 Further `g' tweaks Lars Magne Ingebrigtsen
2010-09-22 22:27 ` Lars Magne Ingebrigtsen
@ 2010-09-23 8:12 ` David Engster
2010-09-23 15:55 ` Lars Magne Ingebrigtsen
1 sibling, 1 reply; 11+ messages in thread
From: David Engster @ 2010-09-23 8:12 UTC (permalink / raw)
To: ding
Lars Magne Ingebrigtsen writes:
> I'm oscillating a bit back and forth between different `g' tweaks, but
> Gnus had gone back to statting thousands of files again, so I did some
> burrowing down and tried to find the reason.
>
> It's back to 200 stats again for me, down from 5K stats earlier this
> evening.
>
> If I broke something again, let me know. :-)
Whatever you have done, it's pretty good. I repeated by benchmark-run of
calling 50 times `gnus-group-get-new-news'. The new code now takes
~12.8secs, while the old code takes ~15secs with my setup (one IMAP
server, 15 groups checked with lots of mail in them).
The new code now does ~60 GCs in those 50 runs, taking about 2.6secs,
which is roughly half of what I've seen before. Still, there's probably
room for improvement. :-)
-David
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Further `g' tweaks
2010-09-23 8:12 ` David Engster
@ 2010-09-23 15:55 ` Lars Magne Ingebrigtsen
2010-09-23 17:25 ` David Engster
0 siblings, 1 reply; 11+ messages in thread
From: Lars Magne Ingebrigtsen @ 2010-09-23 15:55 UTC (permalink / raw)
To: ding
David Engster <deng@randomsample.de> writes:
> Whatever you have done, it's pretty good. I repeated by benchmark-run of
> calling 50 times `gnus-group-get-new-news'. The new code now takes
> ~12.8secs, while the old code takes ~15secs with my setup (one IMAP
> server, 15 groups checked with lots of mail in them).
12 secs? That's way too long, I think. :-)
What does the *gnus trace* buffer look like on a single `g'?
--
(domestic pets only, the antidote for overdose, milk.)
larsi@gnus.org * Lars Magne Ingebrigtsen
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Further `g' tweaks
2010-09-23 15:55 ` Lars Magne Ingebrigtsen
@ 2010-09-23 17:25 ` David Engster
2010-09-23 17:29 ` Lars Magne Ingebrigtsen
0 siblings, 1 reply; 11+ messages in thread
From: David Engster @ 2010-09-23 17:25 UTC (permalink / raw)
To: ding
Lars Magne Ingebrigtsen writes:
> David Engster <deng@randomsample.de> writes:
>
>> Whatever you have done, it's pretty good. I repeated by benchmark-run of
>> calling 50 times `gnus-group-get-new-news'. The new code now takes
>> ~12.8secs, while the old code takes ~15secs with my setup (one IMAP
>> server, 15 groups checked with lots of mail in them).
>
> 12 secs? That's way too long, I think. :-)
Erm... you did read that part with "calling 50 times", right? ;-)
So we're speaking of ~1/4s per get-new-news. I think that's acceptable.
-David
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Further `g' tweaks
2010-09-23 17:25 ` David Engster
@ 2010-09-23 17:29 ` Lars Magne Ingebrigtsen
0 siblings, 0 replies; 11+ messages in thread
From: Lars Magne Ingebrigtsen @ 2010-09-23 17:29 UTC (permalink / raw)
To: ding
David Engster <deng@randomsample.de> writes:
>>> Whatever you have done, it's pretty good. I repeated by benchmark-run of
>>> calling 50 times `gnus-group-get-new-news'. The new code now takes
>>> ~12.8secs, while the old code takes ~15secs with my setup (one IMAP
>>> server, 15 groups checked with lots of mail in them).
>>
>> 12 secs? That's way too long, I think. :-)
>
> Erm... you did read that part with "calling 50 times", right? ;-)
Yes, but I thought you meant that the mean time you got was 12 seconds
per call. :-)
> So we're speaking of ~1/4s per get-new-news. I think that's acceptable.
Yup.
--
(domestic pets only, the antidote for overdose, milk.)
larsi@gnus.org * Lars Magne Ingebrigtsen
^ permalink raw reply [flat|nested] 11+ messages in thread
end of thread, other threads:[~2010-09-23 18:39 UTC | newest]
Thread overview: 11+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2010-09-22 22:06 Further `g' tweaks Lars Magne Ingebrigtsen
2010-09-22 22:27 ` Lars Magne Ingebrigtsen
2010-09-22 23:18 ` Dan Christensen
2010-09-23 0:36 ` Dan Christensen
2010-09-23 15:55 ` Lars Magne Ingebrigtsen
2010-09-23 18:31 ` Dan Christensen
2010-09-23 18:39 ` Lars Magne Ingebrigtsen
2010-09-23 8:12 ` David Engster
2010-09-23 15:55 ` Lars Magne Ingebrigtsen
2010-09-23 17:25 ` David Engster
2010-09-23 17:29 ` Lars Magne Ingebrigtsen
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).