Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On Sat, Nov 10, 2007 at 02:10:39AM +0100, Matteo Tescione wrote: > Leroy, drbd folks, > > After a while of collecting and profiling my systems i discovered that our > poor performance problems were not related of processor, raid controller or > kernel configuration. > It's all about raid5/6 level configurations that is absolutely unreliable. > > In details reading and writing directly to the raid disk, with an ext3 > partition for example, gives more than 200 mbyte/sec with await and svc time > of approx 1.2 ms versus reading and writing with drbd (either syncing or > live i/o) gives extremely poor performance (i got 30-40 mbyte/sec with 3ware > 9650-8p configured in raid6 and i have 20 mbytes/sec with Compaq Smart Array > p400 configured in raid5 mode. > In both raid5/6 configurations, i/o seems stucked as i've reported in my > previous post with very high iowait values (await up to 5000 ms, svctime up > to 1000 ms) > > Switching to raid0 / raid10 configurations, totally solved our problems. > Sync its up to 200mbyte/sec, live i/o seems good as well but i'm still > testing the vmware infrastructure... > > If anyone has any idea of what's going on.... try to put the drbd meta data on a raid1, and leave the data area on raid5 or 6. my guess is that the latency of a synchonous sector write in a raid5/6 set is too high. -- : Lars Ellenberg Tel +43-1-8178292-55 : : LINBIT Information Technologies GmbH Fax +43-1-8178292-82 : : Vivenotgasse 48, A-1120 Vienna/Europe http://www.linbit.com :