Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On Wed, Aug 15, 2007 at 11:16:08AM +0200, Lavender, Ben wrote: > I appreciate all your help on this. I'll spare you the details, but > after running a pile of benchmarks to test the settings you suggested, > the bottleneck appears to be the write speed of logical volume > management (the bare disks write much faster, and drbd takes advantage > of it). So it seems I have to decide if I really want LVM. hm... this sound more like you should up the maximum number of requests in flight, that would be the "nr_requests" in /sys/block/*/queue/ sure, lvm introduces some overhead, but not that much. > One quick note, however: while in the pause-sync mode, if I mess with > the mounted partition, drbd is still throwing out a lot of traffic on > the network. This is pretty counter-intuitive to me; I would expect it > not to be talking with the remote disk at all during pause sync. resync: synced area ,- current syncer "position" * WRITE request |===========>..-----.. not yet synced area .-...---| - in the not yet synced area: clean, does not need to be resynced . in the not yet synced area: dirty, needs to be resynced * application WRITE happens, and gets mirrored right away. once this application write completes, things may look like |============>.-----.. not yet synced area .-.-.---| so syncer may have made some progress, but at the same time the application write was mirrored, and set the corresponding area to "clean", it will be skipped by the resync once it reaches that position. if you "pause" sync, it will do just that: pause the resync. application writes will continue to be mirrored right away. that is intentional, by design. It could be done differently, which may even have its advantages, too. But for now, this is just how it works. -- : Lars Ellenberg Tel +43-1-8178292-0 : : LINBIT Information Technologies GmbH Fax +43-1-8178292-82 : : Vivenotgasse 48, A-1120 Vienna/Europe http://www.linbit.com : __ please use the "List-Reply" function of your email client.