[DRBD-user] drbd 0.7.4 half the speed of 0.6.13

Philipp Reisner philipp.reisner at linbit.com
Wed Sep 22 09:56:03 CEST 2004

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


On Tuesday 21 September 2004 23:22, Josh McAllister wrote:
> First of all... THANK YOU, MAJOR improvement:
> 0: cs:SyncSource st:Secondary/Secondary ld:Consistent
>     ns:7376452 nr:0 dw:0 dr:7377568 al:0 bm:449 lo:0 pe:916 ua:279 ap:0
>         [===>................] sync'ed: 16.6% (36391/43591)M
>         finish: 0:09:26 speed: 65,660 (55,432) K/sec
>
> Based purely on perception (I haven't actually timed full sync with both
> versions -- but I could if it would be useful) though the number reported
> (55,432) is still 30% less than it was with 0.6.13 it seems to be
> performing on par. Could it be that the calculation for the average rate is
> a bit more accurate/different in 0.7.x?
>

Well probabely the attached patch will win back the last 30%. May I ask
you again to verify this, and to post your results to the list.

>
> With that out of the way, a couple more observations on this issue:
>
> 1. With max-buffers/max-epoch @ 256/256 as it was, sync rate was about
> 48MB/sec. 2048/2048 gave about 52MB/sec. 1024/2048 yielded the result
> posted (55MB/sec). This seems more sensible.
>

max-epoch-size does not have any effect on resync performace.
max-buffers should be at least 2048 for such performant hardware I think.
It might be a good idea to have max-epoch-size smaller than max-buffers,
to achive best mirroring (cs = connected) performance.

> 2. I've done some additional performance testing with bonnie... some
> potentially interesting results below. Looks like 0.6.13 is still faster
> overall, and neither is even half as fast as direct device. Is this not an
> accurate measure?
>

~80MB/sec on device
~40MB/sec mirrored via DRBD.

Ever tested the performance of the interconnect ?  Jumbo Frames ?

PS: The previous patch already made it into SVN, so this one stack upon 
    the patch posted by Lars.

-Philipp

-- 
: Dipl-Ing Philipp Reisner                      Tel +43-1-8178292-50 :
: LINBIT Information Technologies GmbH          Fax +43-1-8178292-82 :
: Schönbrunnerstr 244, 1120 Vienna, Austria    http://www.linbit.com :
-------------- next part --------------
A non-text attachment was scrubbed...
Name: get_last_30_percent.diff
Type: text/x-diff
Size: 1446 bytes
Desc: not available
URL: <http://lists.linbit.com/pipermail/drbd-user/attachments/20040922/13e54544/attachment.diff>


More information about the drbd-user mailing list