Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
/ 2006-07-11 11:23:47 +0200
\ Julien WICQUART:
> 2/ Remaining questions
>
> 2.1/ Why do I have difference on reading between
> tests with and without DRBD whereas it is said on the
> site that DRBD has no impact on readings?
maybe drbd "kicks" ("unplugs") the lower level device more often,
or, because of the little housekeeping we have to do,
otherwise change the timing behaviour of requests in a
way that meet the performance requirements of your
storage backend better.
> 2.2/ why the reading per character (Sequential Input Per Chr) with DRBD (both versions) is faster
> than without DRBD ?
> Is it because the node2's disk is better, the read is made on the 2 nodes and the fastest response
> is used?
No.
> However I've read on the first page of the DRBD's site :"Reads are always carried out locally."
That is the case, as long as there is good local data.
(no disk failure, not inconistent/sync target ...)
> 2.3/ I've read in the drbd-announce mailinglist of May that :
> "With drbd-0.7.18 on top of DRBD we got only
> about 70MB/sec read performance, while with that modification
> in 0.7.19 we can get the whole 320MB/sec out of DRBD as well."
> The results between DRBD 0.7.10-4 and DRBD 0.7.19-1 seem a bit similary.
> In which conditions, does the read performance increase like this?
this change in 0.7.18 -> 0.7.19 is relevant for block devices
that like to be used with huge read ahead settings.
obviously, comparing 0.7.10 -> 0.7.19 includes this change.
--
: Lars Ellenberg Tel +43-1-8178292-0 :
: LINBIT Information Technologies GmbH Fax +43-1-8178292-82 :
: Schoenbrunner Str. 244, A-1120 Vienna/Europe http://www.linbit.com :
__
please use the "List-Reply" function of your email client.