Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Hi, On Thu, 26 Sep 2013 17:23:31 +0200 Matthias Teege <matthias-drbdu at mteege.de> wrote: > TL;DR: > - local logical volume -> local logical volume: FAST writing speeds > - local LVM2 logical volume -> local drbd device: SLOW writing speeds > Any ideas what could eat up our performance while writing to a local > drbd device? When you use internal meta-data, the meta-data is at the end of your partition. That means that each write on your drbd device becomes write-long_seek-write-long_seek_back on the backing device. When your drbd volume is only 10GB, that effect doesn't hit you as hard as on a 1TB volume. But a seek is a seek and the penalty hits you on all disks with rotating blatters. (You will not be hit by this effect on an all-ssd setup;-) My advice: - Go for external meta-data on another physical disk, preferable on a ssd. - When there is no other disk available, scratch the 1 of your 10-raid underneath as you currently have a 101 (or 011 or 110 depending on the exact setup). - Then if you are using sync-protocol C, your write is still limited by the network-bandwidth and -roundtrip-latency. If that is not enough, you might consider using protocol A. The docs and the experts tell you that you should only use that with battery-backed-controllers. But when your two machines have independant UPS and are either directly connected on the network or with a switch also on UPS, the chance of your second node not 'getting' the write when the first node fails is very small... Have fun, Arnold -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 198 bytes Desc: not available URL: <http://lists.linbit.com/pipermail/drbd-user/attachments/20130928/fc5f4dc3/attachment.pgp>