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 05 October 2010 15:31:56 J. Ryan Earl wrote:
> > Does your RAID controller have a power-backed write
> >
> > > cache?
> >
> > Yes
>
> Then have you set:
>
> disk {
> [...]
> no-disk-barrier;
> no-disk-flushes;
> no-disk-drain;
> no-md-flushes;
> }
Now I will :-)
> > > Have you tried RAID10?
> >
> > No, but since the bonnie++ test without DRBD give a 250 MB/sec
> > performance hit, I guess this is not where our bottleneck is ...
>
> Is your metadata internal? Without DRBD you don't have DRBD's metadata.
> If the metadata is being written with FUA (Force Unit Access) method,
> sometimes it skips even battery-backed caches and issues an write. Which
> on RAID<456> this equals additional reads/writes on drives to complete the
> strip line. Such contention can lower performance. This is common with
> all journals and transaction logs, but it depends on your hardware's FUA
> implementation.
>
> -JR
The metadata is indeed internal. I need to look into the FUA part of the
controller specs, but I will start by implementing your other ideas first.
Since netperf gives me almost the theoretical maximum and the backend storage
scores highly, I suspect the culprit will mostly be a poorly configured
drbd.conf. Next I will attack TCP tuning.
Will report back on this list on Thursday.
Thx again!
B.