You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Neels J Hofmeyr <ne...@elego.de> on 2011/10/31 03:47:56 UTC
trunk is faster. -- Re: [svnbench] Revision: 1195287 compiled Oct
31 2011, 00:21:33
I've reversed the way the timings are printed -- so that the "most
interesting" stuff is at the top. The detail increases the further you read
down.
Across the board these results suggest a tendency for faster operation in
trunk, compared against 1.7.x. 'svn copy' is the only operation with a
slight speed dip (~ 4% slower in trunk).
Trunk's "svn info" shows a near 10-fold speed increase (88% less runtime!)
[1a], and 20% less runtime for 'svn merge' [2a], both in WCs with evenly
spread dirs and files.
Switch, update and commit aren't too bad either.
Most perf improvement is in WCs with evenly spread dirs and files, compare
[1a] [2a] with [1b] [1c] [2b] [2c].
Some more legend:
5x5: 5 dirs and files per dir, 5 levels deep [1a] [2a]
100x1: 100 nested dirs [1b] [2b]
1x100: 100 files in a single dir [1c] [2c]
All this is ra_local only, on GNU/linux.
On 10/31/2011 02:52 AM, neels@apache.org wrote:
> /home/neels/svnbench/20111031-002440
> Started at Mon Oct 31 00:24:40 UTC 2011
>
> Averaged-total results across all runs:
> ---------------------------------------
>
> COMPARE total_1.7.x to total_trunk
> TOTAL RUN timings: 164.1 seconds avg for total_1.7.x
> 128.0 seconds avg for total_trunk
> avg operation
> 0.78|-36.176 TOTAL RUN
> 0.89| -0.002 add
> 0.92| -0.077 checkout
> 0.98| -0.221 commit
> 1.04| +0.008 copy
> 0.86| -0.051 delete
> 0.13| -5.371 info
> 0.81| -0.979 merge
> 0.72| -0.004 mkdir
> 0.85| -0.001 prop mod
> 0.83| -0.002 propdel
> 0.84| -0.001 proplist
> 0.86| -0.001 propset
> 0.83| -0.001 resolve
> 0.92| -0.017 resolved
> 0.94| -0.012 status
> 0.86| -0.200 switch
> 0.90| -0.079 update
> (legend: "1.23|+0.45" means: slower by factor 1.23 and by 0.45 seconds;
> factor < 1 and difference < 0 means 'total_trunk' is faster than 'total_1.7.x')
>
>
> Above totals split into separate <dir-levels>x<dir-spread> runs:
> ----------------------------------------------------------------
>
> COMPARE 5x5_1.7.x to 5x5_trunk
> TOTAL RUN timings: 457.3 seconds avg for 5x5_1.7.x
> 351.7 seconds avg for 5x5_trunk
> avg operation
> 0.77|-105.559 TOTAL RUN
> 0.90| -0.002 add
> 0.92| -0.209 checkout
> 0.98| -0.602 commit
> 1.06| +0.014 copy
> 0.85| -0.137 delete
> 0.12|-15.994 info
^^^^ [1a]
> 0.80| -2.834 merge
^^^^ [2a]
> 0.71| -0.004 mkdir
> 0.85| -0.001 prop mod
> 0.83| -0.002 propdel
> 0.84| -0.001 proplist
> 0.86| -0.001 propset
> 0.85| -0.001 resolve
> 0.93| -0.044 resolved
> 0.95| -0.027 status
> 0.86| -0.537 switch
> 0.90| -0.203 update
> (legend: "1.23|+0.45" means: slower by factor 1.23 and by 0.45 seconds;
> factor < 1 and difference < 0 means '5x5_trunk' is faster than '5x5_1.7.x')
>
> COMPARE 100x1_1.7.x to 100x1_trunk
> TOTAL RUN timings: 26.8 seconds avg for 100x1_1.7.x
> 24.3 seconds avg for 100x1_trunk
> avg operation
> 0.91| -2.479 TOTAL RUN
> 0.87| -0.002 add
> 0.91| -0.018 checkout
> 0.97| -0.044 commit
> 1.01| +0.002 copy
> 0.89| -0.012 delete
> 0.53| -0.120 info
^^^^ [1b]
> 0.88| -0.094 merge
^^^^ [2b]
> 0.85| -0.002 mkdir
> 0.85| -0.002 prop mod
> 0.86| -0.002 propdel
> 0.85| -0.002 proplist
> 0.86| -0.002 propset
> 0.85| -0.001 resolve
> 0.91| -0.007 resolved
> 0.92| -0.006 status
> 0.87| -0.053 switch
> 0.89| -0.029 update
> (legend: "1.23|+0.45" means: slower by factor 1.23 and by 0.45 seconds;
> factor < 1 and difference < 0 means '100x1_trunk' is faster than '100x1_1.7.x')
>
> COMPARE 1x100_1.7.x to 1x100_trunk
> TOTAL RUN timings: 8.3 seconds avg for 1x100_1.7.x
> 7.8 seconds avg for 1x100_trunk
> avg operation
> 0.94| -0.491 TOTAL RUN
> 0.92| -0.003 add
> 0.95| -0.004 checkout
> 0.97| -0.016 commit
> 1.04| +0.007 copy
> 0.88| -0.003 delete
> 1.00| -0.000 info
^^^^ [1c]
> 0.95| -0.010 merge
^^^^ [2c]
> 0.85| -0.001 prop mod
> 0.86| -0.001 propdel
> 0.84| -0.001 proplist
> 0.85| -0.001 propset
> 0.80| -0.002 resolve
> 0.88| -0.002 resolved
> 0.90| -0.002 status
> 0.89| -0.009 switch
> 0.88| -0.005 update
> (legend: "1.23|+0.45" means: slower by factor 1.23 and by 0.45 seconds;
> factor < 1 and difference < 0 means '1x100_trunk' is faster than '1x100_1.7.x')
>
>
>
> More detail:
> ------------
[...]
~Neels
Re: trunk is faster. -- Re: [svnbench] Revision: 1195287 compiled Oct
31 2011, 00:21:33
Posted by Daniel Shahaf <d....@daniel.shahaf.name>.
On Tuesday, November 01, 2011 1:18 PM, "Neels J Hofmeyr" <ne...@elego.de> wrote:
> On 10/31/2011 05:15 AM, Daniel Shahaf wrote:
> > On Monday, October 31, 2011 3:47 AM, "Neels J Hofmeyr" <ne...@elego.de> wrote:
> >> All this is ra_local only, on GNU/linux.
> >
> > on a VMware VM.
>
> indeed.
> The test does run both builds alternately, so it does provide some measure
> of difference between the two builds (1.7.x and trunk).
>
> > We also have a FreeBSD jail sitting around and, AFAICT, doing nothing. We could run the tests on it too if it helps our work and doesn't negatively affect the sibling jails.
>
> I don't really care, but np if you want me to...
>
> ~Neels
Thanks, but I don't have a strong opinion either --- I was just surveying our options, rather than recommending any one of them.
Re: trunk is faster. -- Re: [svnbench] Revision: 1195287 compiled
Oct 31 2011, 00:21:33
Posted by Neels J Hofmeyr <ne...@elego.de>.
On 10/31/2011 05:15 AM, Daniel Shahaf wrote:
> On Monday, October 31, 2011 3:47 AM, "Neels J Hofmeyr" <ne...@elego.de> wrote:
>> All this is ra_local only, on GNU/linux.
>
> on a VMware VM.
indeed.
The test does run both builds alternately, so it does provide some measure
of difference between the two builds (1.7.x and trunk).
> We also have a FreeBSD jail sitting around and, AFAICT, doing nothing. We could run the tests on it too if it helps our work and doesn't negatively affect the sibling jails.
I don't really care, but np if you want me to...
~Neels
Re: trunk is faster. -- Re: [svnbench] Revision: 1195287 compiled Oct
31 2011, 00:21:33
Posted by Daniel Shahaf <d....@daniel.shahaf.name>.
On Monday, October 31, 2011 3:47 AM, "Neels J Hofmeyr" <ne...@elego.de> wrote:
> All this is ra_local only, on GNU/linux.
on a VMware VM.
We also have a FreeBSD jail sitting around and, AFAICT, doing nothing. We could run the tests on it too if it helps our work and doesn't negatively affect the sibling jails.