monotone-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Monotone-devel] Re: linus talk on git


From: Benoît Dejean
Subject: Re: [Monotone-devel] Re: linus talk on git
Date: Tue, 29 May 2007 21:23:33 +0200

Le mardi 29 mai 2007 à 13:34 +0100, Bruce Stephens a écrit :
> Nathaniel Smith <address@hidden> writes:
> 
> [...]
> 
> > Taking 3.5s to stat Derek's directory tree is a straight bug too, is
> > my point.
> 
> Oh, yes.  At least for an unchanged tree, there's no reason monotone
> has to be slower than git for this, since the inodeprints file is just
> an ordinary local file.  And I guess it's possible the bug is indeed
> related: what I was seeing was just *so* slow that when it got
> reasonable I assumed the problem had been fixed, but perhaps it's
> doing n times as many stats as necessary for some smaller n.  Should
> be easy to test: count the number of stats for a couple of different
> sized checkouts.

Almost 1s here, local fs :

/usr/bin/time strace -c mtn status
Current branch: net.venge.monotone
Changes against parent 869baba046ea547aadee135aec6938e2f79d3c50
  patched  file_io.cc
  patched  file_io.hh
  patched  work.cc
Process 17036 detached
% time     seconds  usecs/call     calls    errors syscall
------ ----------- ----------- --------- --------- ----------------
 62.90    0.006666          19       355           read
 31.44    0.003332          85        39           mmap
  4.01    0.000425           0      3928         1 stat64
  1.00    0.000106           3        31           brk
  0.33    0.000035           0      1670           time
  0.31    0.000033           1        29        27 access
[...]
------ ----------- ----------- --------- --------- ----------------
100.00    0.010597                  6475        34 total
0.85user 0.17system 0:01.10elapsed 93%CPU (0avgtext+0avgdata
0maxresident)k
0inputs+0outputs (0major+2551minor)pagefaults 0swaps

It looks CPU bounded, so i don't think. In the other thread i said
that /2 the number of stat calls doesn't improve overall time on a local
filesystem.

-- 
Benoît Dejean
GNOME http://www.gnomefr.org/
LibGTop http://directory.fsf.org/libgtop.html

Attachment: signature.asc
Description: Ceci est une partie de message numériquement signée


reply via email to

[Prev in Thread] Current Thread [Next in Thread]