Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On Saturday 18 October 2008 21:18, Petersen, Joerg wrote: > Well, my experience is that 8.0.7 is the fastest DRDB... > Later Versions just got slower! > Try: no-disk-flushes > and: no-md-flushes > if you are using DRBD >= 8.0.13 > Not completely same to 8.0.7 but near to it... > > Mit freundlichen Grüßen > Jörg > > -----Ursprüngliche Nachricht----- > Von: drbd-user-bounces at lists.linbit.com > [mailto:drbd-user-bounces at lists.linbit.com] Im Auftrag von Bart Coninckx > Gesendet: Freitag, 17. Oktober 2008 16:40 > An: drbd-user at lists.linbit.com > Betreff: [DRBD-user] Re: drbd is performing at 12 MB/sec on recent > > > On Thu, Oct 16, 2008 at 10:19:20AM +0200, Bart Coninckx wrote: > > > > fix your local io subsystem. > > > > fix your local io subsystem driver. > > > > > > > > check > > > > local io subsystem, > > > > (battery backed) write cache enabled? > > > > > > Yes, I did that yesterday, I got from 13 MB/Sec to 16 MB/Sec. > > > Better, but still not good. > > > > > > > raid resync/rebuild/re-whatever going on at the same time? > > > > > > nope, RAID has been built and is steady. > > > > > > > does the driver honor BIO_RW_SYNC? > > > > > > errr ... how can I check that? /proc/driver/cciss/cciss0 does not > > > reveal anything about that. > > > > > > > does the driver introduce additional latency because of > > > > ill-advised "optimizations"? > > > > > > I use a stock SLES 10 SP1 cciss driver. I did understand that the > > > I/O scheduler is not the best one. Perhaps you are referring to that? > > > > > > > if local io subsystem is ok, > > > > but DRBD costs more than a few percent in throughput, check your > > > > local io subsystem on the other node! > > > > > > That node is identical to the other one. > > > > > > > if that is ok as well, check network for throughput, latency and > > > > packet loss/retransmits. > > > > > > scp gives me about 70 MB/Sec, so my guess is that things are more > > > likely related to drbd and local io. > > > > scp does not know about fsync. > > > > use the same benchmark against drbd and non-drbd partition on that > > cciss... > > Hi Lars, > > I did these: > > node1:/opt # hdparm -t /dev/drbd0 > > /dev/drbd0: > Timing buffered disk reads: 1016 MB in 3.00 seconds = 338.34 MB/sec > > node1:/opt # hdparm -t /dev/mapper/vg1-opt > > /dev/mapper/vg1-opt: > Timing buffered disk reads: 730 MB in 3.01 seconds = 242.79 MB/sec > > node1:/opt # hdparm -t /dev/cciss/c0d0p3 > > /dev/cciss/c0d0p3: > Timing buffered disk reads: 1666 MB in 3.01 seconds = 554.10 MB/sec > > > The first one is obviously drbd (on top of LVM) The second one is /opt in > the same LVM group The third is the root partition outside of LVM. > > > This would suggest LVM slows things somewhat down. However, I did setup > drbd outside of LVM and this gave no different sync results (did no > performance tests once synced, since this the syncing took so long). > > Do these tests make any sence? > > > thx! > > > Bart Hi Joerg, I 'm using drbd-0.7.22, since I've installed SLES 10 SP1. As you probably know, one looses support for software versions other than those supplied with the DVDs. Rgds, Bart