[DRBD-user] Slow Read performance

[Maxim Ianoglo]

Hi,

> I'm not familiar with this tool. Does it mix reads and writes?
No it does not mix them.
Here is it's testing logic ( seems to be ):
Writing a byte at a time...done
Writing intelligently...done
Rewriting...done
Reading a byte at a time...done
Reading intelligently...done
start 'em...done...done...done...done...done...
Create files in sequential order...done.
Stat files in sequential order...done.
Delete files in sequential order...done.
Create files in random order...done.
Stat files in random order...done.
Delete files in random order...done.

> You may want to consider activating disk flushes. Why does your config
disable those?
Because of BBU present on RAID controller.
http://www.drbd.org/users-guide/s-throughput-tuning.html

> Looking at your write throughput being much larger than your read
> throuhgput(!) I suspect you're writing to some sort of aggressive cache,
> and your reads must wait until the writes are completed to the disk(?)
> If that is indeed the case, this might explain why syncing to the
> secondary slows down the read performance.
I noticed this. Could be "feature" of H700 RAID controller. Anyway I have
changed Write Cache policy to
Write-Through, so theoretically I would have more "real" results. But I
still get same results. I think I will try to contact DELL support/forum ...

Thank you.
--
Maxim Ianoglo

On Mon, May 2, 2011 at 11:05 AM, Felix Frank <ff at mpexnet.de> wrote:

> Hi,
>
> On 04/30/2011 11:12 PM, Maxim Ianoglo wrote:
> > Hello,
> >
> > Have two DELL PE R410, RAID 10 15K rpm Seagate SAS disks. Both servers
> > are connected via 1Gb link.
> >
> > Testing performance with bonnie++.
> > On a RAW device  I get: ~240MB/s Read and ~390MB/s Write
> > On DRBD device with Secondary DRBD host connected: ~80MB/s Read and
> > ~340MB/s Write
> > On DRBD without Secondary DRBD host connected: ~245MB/s Read and
> ~340Write
>
> I'm not familiar with this tool. Does it mix reads and writes?
>
> Looking at your write throughput being much larger than your read
> throuhgput(!) I suspect you're writing to some sort of aggressive cache,
> and your reads must wait until the writes are completed to the disk(?)
> If that is indeed the case, this might explain why syncing to the
> secondary slows down the read performance.
> This seems especially likely since your write throughput is
> a) not affected by live syncing and
> b) WAY above the capacity of a 1GB link!
>
> Long story short, I disbelieve your mode of testing is yielding results
> of any operational value.
>
> Also note that a volatile write cache may leave your secondary (and
> primary) disks in a corrupted state on primary crash (and the secondary
> even during connection loss). You may want to consider activating disk
> flushes. Why does your config disable those?
>
> > Why I get so bad Read performance on DRBD device with Secondary host
> > connected ? Per documentation all reads should be made l
>
> Yes they are local.
>
> Also note:
>
> >   syncer {
> >     cpu-mask 4095;
> >     rate 85M;
> >     verify-alg "crc32c";
> >     csums-alg "crc32c";
> >   }
>
> It is advisable to not use syncer rates that exceed 30% of your
> bandwidth by much.
>
> This rate is only used if an outdated peer is doing a full or partial
> sync, it doesn't affect live synchronization (which is always done as
> soon as each block is written).
> If you use up most of your bandwidth for syncing, the write performance
> on your DRBD will suffer.
>
> HTH,
> Felix
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linbit.com/pipermail/drbd-user/attachments/20110502/5923cb1a/attachment.htm>