git.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Brandon Casey <casey@nrlssc.navy.mil>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Alex Riesen <raa.lkml@gmail.com>,
	Bevan Watkiss <bevan.watkiss@cloakware.com>,
	Git Mailing List <git@vger.kernel.org>
Subject: Re:
Date: Fri, 08 May 2009 12:27:05 -0500	[thread overview]
Message-ID: <Ah7lj3UWxgwxNiQs6kqiiVurulv4F00ssWrb3OzfTrXYlK8ZBCSBOQ@cipher.nrlssc.navy.mil> (raw)
In-Reply-To: <alpine.LFD.2.01.0905080857130.4983@localhost.localdomain>

Linus Torvalds wrote:
> 
> On Fri, 8 May 2009, Brandon Casey wrote:
>> plain 'git checkout' on linux kernel over NFS.
> 
> Thanks.
> 
>> Best time without patch: 1.20 seconds
>>
>>   0.45user 0.71system 0:01.20elapsed 96%CPU (0avgtext+0avgdata 0maxresident)k
>>   0inputs+0outputs (0major+15467minor)pagefaults 0swaps
>>
>> Best time with patch (core.preloadindex = true): 1.10 seconds
>>
>>   0.43user 4.00system 0:01.10elapsed 402%CPU (0avgtext+0avgdata 0maxresident)k
>>   0inputs+0outputs (0major+13999minor)pagefaults 0swaps
>>
>> Best time with patch (core.preloadindex = false): 0.84 seconds
>>
>>   0.42user 0.39system 0:00.84elapsed 96%CPU (0avgtext+0avgdata 0maxresident)k
>>   0inputs+0outputs (0major+13965minor)pagefaults 0swaps
> 
> Ok, that is _disgusting_. The parallelism clearly works (402%CPU), but the 
> system time overhead is horrible. Going from 0.39s system time to 4s of 
> system time is really quite nasty.
> 
> Is there any possibility you could oprofile this (run it in a loop to get 
> better profiles)? It very much sounds like some serious lock contention, 
> and I'd love to hear more about exactly which lock it's hitting.

Possibly, I'll see if our sysadmin has time to "play".

> Also, you're already almost totally CPU-bound, with 96% CPU for the 
> single-threaded csase. So you may be running over NFS, but your NFS server 
> is likely pretty good and/or the client just captures everything in the 
> caches anyway.
> 
> I don't recall what the Linux NFS stat cache timeout is, but it's less 
> than a minute. I suspect that you ran things in a tight loop, which is why 
> you then got effectively the local caching behavior for the best times. 

Yeah, that's what I did.

> Can you do a "best time" check but with a 60-second pause between runs 
> (and before), to see what happens when the client doesn't do caching?

No problem.

>> Best time with read_cache_preload patch only: 1.38 seconds
>>
>>   0.45user 4.42system 0:01.38elapsed 352%CPU (0avgtext+0avgdata 0maxresident)k
>>   0inputs+0outputs (0major+13990minor)pagefaults 0swaps
> 
> Yeah, here you're not getting any advantage of fewer lstats, and you 
> show the same "almost entirely CPU-bound on four cores" behavior, and the 
> same (probable) lock contention that has pushed the system time way up.
> 
>> The read_cache_preload() changes actually slow things down for me for this
>> case.
>>
>> Reduction in lstat's gives a nice 30% improvement.
> 
> Yes, I think the one-liner lstat avoidance is a real fix regardless. And 
> the preloading sounds like it hits serialization overhead in the kernel, 
> which I'm not at all surprised at, but not being surprised doesn't mean 
> that I'm not interested to hear where it is.
> 
> The Linux VFS dcache itself should scale better than that (but who knows - 
> cacheline ping-pong due to lock contention can easily cause a 10x slowdown 
> even without being _totally_ contended all the time). So I would _suspect_ 
> that it's some NFS lock that you're seeing, but I'd love to know more.
> 
> Btw, those system times are pretty high to begin with, so I'd love to know 
> kernel version and see a profile even without the parallel case and 
> presumably lock contention. Because while I probably have a faster 
> machine anyway, what I see iis:
> 
> 	[torvalds@nehalem linux]$ /usr/bin/time git checkout
> 	0.13user 0.05system 0:00.19elapsed 98%CPU (0avgtext+0avgdata 0maxresident)k
> 	0inputs+0outputs (0major+13334minor)pagefaults 0swaps
> 
> ie my "system" time is _much_ lower than yours (and lower than your system 
> time). This is the 'without patch' time, btw, so this has extra lstat's. 
> And my system time is still lower than my user time, so I wonder where all 
> _your_ system time comes from. Your system time is much more comparable to 
> user time even in the good case, and I wonder why?
> 
> Could be just that kernel code tends to have more cache misses, and my 8MB 
> cache captures it all, and yours doesn't. Regardless, a profile would be 
> very interesting.

Something is definitely up.

I provided timing results for your original preload_cache implementation
which affected status and diff, which was part of the justification for
merging it in.

   http://article.gmane.org/gmane.comp.version-control.git/100998

You can see that cold cache system time for 'git status' went from 0.36 to
0.52 seconds.  Fine.  I just ran it again, and now I'm getting system time
of 10 seconds!  This is the same machine.

Similarly for the cold cache 'git checkout' reruns:

Best without patch: 6.02 (systime 1.57)

  0.43user 1.57system 0:06.02elapsed 33%CPU (0avgtext+0avgdata 0maxresident)k
  5336inputs+0outputs (12major+15472minor)pagefaults 0swaps

Best with patch (preload_cache,lstat reduction): 2.69 (systime 10.47)

  0.45user 10.47system 0:02.69elapsed 405%CPU (0avgtext+0avgdata 0maxresident)k
  5336inputs+0outputs (12major+13985minor)pagefaults 0swaps


OS: Centos4.7

$ cat /proc/version
Linux version 2.6.9-78.0.17.ELsmp (mockbuild@builder16.centos.org) (gcc version 3.4.6 20060404 (Red Hat 3.4.6-9)) #1 SMP Thu Mar 12 20:05:15 EDT 2009

-brandon

  reply	other threads:[~2009-05-08 17:28 UTC|newest]

Thread overview: 83+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2009-05-07 17:01 (unknown), Bevan Watkiss
2009-05-07 17:13 ` Alex Riesen
2009-05-07 17:26   ` Bevan Watkiss
2009-05-07 18:18     ` Alex Riesen
2009-05-07 18:48       ` Bevan Watkiss
2009-05-07 19:56         ` Björn Steinbrink
2009-05-07 18:56     ` Linus Torvalds
2009-05-07 19:37       ` RE: Bevan Watkiss
2009-05-07 20:07         ` RE: Linus Torvalds
2009-05-07 20:20           ` RE: Linus Torvalds
2009-05-07 20:43             ` Junio C Hamano
2009-05-07 21:33               ` Re: Linus Torvalds
2009-05-07 21:55             ` Linus Torvalds
2009-05-07 22:27               ` RE: david
2009-05-07 22:36                 ` RE: Linus Torvalds
2009-05-07 22:43                   ` RE: david
2009-05-07 23:00                     ` RE: Linus Torvalds
2009-05-07 23:07                       ` RE: david
2009-05-07 23:18                         ` RE: Linus Torvalds
2009-05-07 23:31                           ` RE: david
2009-05-07 23:57                             ` Johan Herland
2009-05-08 16:14                             ` Bevan Watkiss
2009-05-08  8:17               ` Alex Riesen
2009-05-08 14:39                 ` Re: Linus Torvalds
2009-05-08 15:51                   ` Re: Brandon Casey
2009-05-08 16:15                     ` Re: Linus Torvalds
2009-05-08 17:27                       ` Brandon Casey [this message]
2009-05-08 17:43                         ` Re: Brandon Casey
2009-05-08 21:49                           ` Re: Linus Torvalds
2009-05-08 23:04                             ` Re: Brandon Casey
2009-05-09 16:44                               ` Re: Linus Torvalds
2009-05-08 17:44                         ` Re: Linus Torvalds
2009-05-08 16:47               ` 'git checkout' and unlink() calls (was: Re: ) Kjetil Barvik
2009-05-08 17:57                 ` Linus Torvalds
     [not found] <20220301070226.2477769-1-jaydeepjd.8914>
2022-03-06 11:10 ` Jaydeep P Das
2022-03-06 11:22   ` Jaydeep Das
  -- strict thread matches above, loose matches on Subject: below --
2021-08-21 14:40 TECOB270_Ganesh Pawar
2021-08-21 23:52 ` Jeff King
2019-11-15 16:03 Martin Nicolay
2019-11-15 16:29 ` Martin Ågren
2019-11-15 16:37   ` Re: Martin Ågren
2019-08-20 17:23 William Baker
2019-08-20 17:27 ` Yagnatinsky, Mark
2019-03-05 14:57 [GSoC][PATCH v2 3/3] t3600: use helpers to replace test -d/f/e/s <path> Eric Sunshine
2019-03-05 23:38 ` Rohit Ashiwal
2019-01-23 10:50 Christopher Hagler
2019-01-23 14:16 ` Cody Kratzer
2019-01-23 14:25   ` Re: Thomas Braun
2019-01-23 16:00   ` Re: Christopher Hagler
2019-01-23 16:35     ` Randall S. Becker
2019-01-24 17:11   ` Johannes Schindelin
2018-10-08 13:33 Netravnen
2018-10-08 13:34 ` Inderpreet Saini
2018-04-27  0:54 [PATCH v3 2/3] merge: Add merge.renames config setting Ben Peart
2018-04-27 18:19 ` Elijah Newren
2018-04-30 13:11   ` Ben Peart
2018-04-30 16:12     ` Re: Elijah Newren
2018-05-02 14:33       ` Re: Ben Peart
2018-02-27  1:18 Alan Gage
2018-02-27 10:26 ` René Scharfe
2017-11-20 15:10 Viet Nguyen
2017-11-20 20:07 ` Stefan Beller
2017-11-12  2:21 hsed
2017-11-13 18:56 ` Stefan Beller
2017-01-25  0:11 [PATCH 7/7] completion: recognize more long-options Cornelius Weig
2017-01-25  0:21 ` Stefan Beller
2017-01-25  0:43   ` Cornelius Weig
2017-01-25  0:52     ` Re: Stefan Beller
2017-01-25  0:54   ` Re: Linus Torvalds
2017-01-25  1:32     ` Re: Eric Wong
2016-04-11 19:04 (unknown), miwilliams
2016-04-12  4:33 ` Stefan Beller
2015-08-19 19:41 Re: christain147
2015-08-19 11:09 Re: christain147
2015-08-05 12:47 (unknown) Ivan Chernyavsky
2015-08-15  9:19 ` Duy Nguyen
2015-08-17 17:49   ` Re: Junio C Hamano
2015-04-08 20:44 (unknown), Mamta Upadhyay
2015-04-08 21:58 ` Thomas Braun
2015-04-09 11:27   ` Re: Konstantin Khomoutov
     [not found] <CANSxx61FaNp5SBXJ8Y+pWn0eDcunmibKR5g8rttnWGdGwEMHCA@mail.gmail.com>
2015-03-18 20:45 ` Re: Junio C Hamano
2015-03-18 21:06   ` Re: Stefan Beller
2015-03-18 21:17     ` Re: Jeff King
2015-03-18 21:28       ` Re: Jeff King
2015-03-18 21:33         ` Re: Junio C Hamano
2015-03-18 21:45           ` Re: Stefan Beller
2015-03-13  1:34 (unknown) cody.taylor
2015-03-13  2:00 ` Duy Nguyen
2014-09-08 11:36 (unknown), R. Klomp
     [not found] ` <CAOqJoqGSRUw_UT4LhqpYX-WX6AEd2ReAWjgNS76Cra-SMKw3NQ@mail.gmail.com>
2014-09-08 14:36   ` R. Klomp
2014-09-10  0:00     ` Re: David Aguilar
2014-09-15 15:10       ` Re: R. Klomp
2014-02-06 11:54 "Sparse checkout leaves no entry on working directory" all the time on Windows 7 on Git 1.8.5.2.msysgit.0 konstunn
2014-02-06 13:20 ` Johannes Sixt
2014-02-06 19:56   ` Constantine Gorbunov
2012-06-12 21:12 (unknown), rohit sood
2012-06-12 23:51 ` Erik Faye-Lund
2009-11-18  5:03 Re: Anna
2009-05-11 18:57 (unknown) Don Slutz
2009-05-11 20:48 ` Johannes Schindelin
2009-05-12 12:45   ` Re: Don Slutz
2009-03-30  5:03 (unknown), David Aguilar
2009-03-30  7:02 ` Markus Heidelberg
2009-03-30  8:46   ` Re: Junio C Hamano
2007-11-01 20:44 (unknown), Francesco Pretto
2007-11-01 20:48 ` Francesco Pretto
2006-02-02  0:39 [RFC & PATCH] Solaris 8: ENOSYS when mkdir applied to automount., Jason Riedy
2006-02-02  4:18 ` H. Peter Anvin
2005-04-22 22:19 (unknown), atani
2005-04-22 23:16 ` Martin Schlemmer

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=Ah7lj3UWxgwxNiQs6kqiiVurulv4F00ssWrb3OzfTrXYlK8ZBCSBOQ@cipher.nrlssc.navy.mil \
    --to=casey@nrlssc.navy.mil \
    --cc=bevan.watkiss@cloakware.com \
    --cc=git@vger.kernel.org \
    --cc=raa.lkml@gmail.com \
    --cc=torvalds@linux-foundation.org \
    --subject='Re:' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

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).