Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Hello, I am running a 2-node active/passive DRBD cluster with DRBD running on top of LVM which itself is on top of a 10x2TB software RAID of 7200RPM SATAII HDDs. Both nodes are running Ubuntu 12.04 amd64. The DRBD replication link is a direct 10Gb SFP connection with the MTU increased to 9000. Here is a link to the DRBD configuration: http://pastebin.com/AQuNYN5s I am concerned in general about the poor write performance that I am seeing, however in particular I've recently noticed odd behavior where the secondary (passive) node's disk utilization is much higher than the active node (using iostat): # iostat -dxmN 5: on primary node: Device: rrqm/s wrqm/s r/s w/s rMB/s wMB/s avgrq-sz avgqu-sz await r_await w_await svctm %util storage-drives 0.00 0.00 12.40 447.80 0.05 1.78 8.17 2.51 5.46 20.26 5.05 0.54 25.04 on secondary node: Device: rrqm/s wrqm/s r/s w/s rMB/s wMB/s avgrq-sz avgqu-sz await r_await w_await svctm %util storage-drives 0.00 0.00 0.00 524.60 0.00 1.80 7.02 3.17 6.05 0.00 6.05 1.76 92.08 I started iostat on both nodes at approximately the same time (hence the slight variation in columns like wMB/s), however the utilization variation by around 50% between the nodes. They are both currently UpToDate so a resync cannot be the explanation for this difference. Write speeds on small files are very slow. Here are some tests I did with dd while adjusting various settings recommended by the DRBD Users Guide (e.g. deadline scheduler with front_merges, read_expire, and write_expire adjusted): http://pastebin.com/64xn2ptg Writing out 100KB files results in around 7MB/s write performance. Writing 10MB files results in better performance, around 40MB/s. What can I do to increase performance, particularly with small files? Thanks, Andrew