Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
/ 2005-07-28 09:58:08 +1000 \ Bradley Baetz: > On Thu, Jul 28, 2005 at 09:12:49AM +1000, Bradley Baetz wrote: > > I'm trying to test the 0.7.11 version of drbd (upgrading from 0.6 on a > > 2.4 kernel). > > > Then on giveback, the system started doing a full sync: > > Theres something else odd going on here, too. During the resync, if the > SyncTarget is Primary, I get a _lot_ better performance than if the > SyncTarget is Secondary. > > 'tested' with a quick and dirty pgbench run. Just dd'ing 1G from > /dev/zero into a file (and then syncing) takes 4min6sec when the SyncTarget > is primary, vs 4min45sec when Secondary. this is repeatable, and I didn't > see this performance issue when not doing a resync. > > What would cause this? I'm running drbd on top of LVM, if that matters. often the more "intelligent" storage devices try to be smart on you and optimize for higher throughput causing more latency, if the controler is not saturated. obviously on a sync target, there are more requests, and the controller is more likely to be saturated with requests. also seek timing may be different. if you are still on 2.4 kernel, you have just one global tq_disk, that may have an effect, too. and there are more complex internal locking and coherency issues which may cause drbd to perform better during sync if the sync target is primary. but you want to stay away from a Primary SyncTarget if you can, since if the network has a hickup, your primary will panic with "LOST... sorry, no access to good data any more" or some such. this state is only for those that need a more graceful failover in case of local io error than a panic triggered one. and, unfortunately, panic does not always succeed in really killing/rebooting the box, especcially on SMP systems, or if reiserfs is involved (thats just from personal experience). -- : 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.