Description of problem: ------------------------ A small regression was introduced with 3.7.9-4 on small file mkdirs. Benchmark with 3.1.2 : 2716.356667 files/sec What I'm seeing on 3.7.9-4 is ~2200-2300 files/sec This is without any performance tuning/enhancement. More on workload details / volume profiles later. Version-Release number of selected component (if applicable): ------------------------------------------------------------- glusterfs-3.7.9-4.el6rhs.x86_64 How reproducible: ---------------- 100% Steps to Reproduce: ------------------ 1. Establish baseline for mkdir on 3.1.2. 2. Run the same test on 3.1.3 3. Check for any regression.Should be within +-10%. Actual results: --------------- The operation has regressed by ~17% Expected results: ---------------- The results should be within 10% of the baseline. Additional info: --------------- > CONFIG : *OS* : RHEL 6.X [root@gqas013 ~]# gluster v info testvol Volume Name: testvol Type: Distributed-Replicate Volume ID: 2930ec4a-5f3a-4ac2-a3fc-5b792ef29383 Status: Started Number of Bricks: 2 x 2 = 4 Transport-type: tcp Bricks: Brick1: gqas013.sbu.lab.eng.bos.redhat.com:/bricks/testvol_brick5 Brick2: gqas011.sbu.lab.eng.bos.redhat.com:/bricks/testvol_brick1 Brick3: gqas005.sbu.lab.eng.bos.redhat.com:/bricks/testvol_brick2 Brick4: gqas006.sbu.lab.eng.bos.redhat.com:/bricks/testvol_brick3 Options Reconfigured: server.allow-insecure: on performance.stat-prefetch: off performance.readdir-ahead: on [root@gqas013 ~]# > rmdir regression is tracked via https://bugzilla.redhat.com/show_bug.cgi?id=1330235
Since the dev_freeze criteria is met, meeting the blocker criteria requires at least 20% drop in performance. So this doesn't qualify as a blocker. Can we have an agreement on that and based on it clear the blocker flag. However dev will initiate an analysis to see what has caused this performance drop. Assigning it to Raghavendra.
Karan/Ambarish, Though I am skeptical that [1] improves performance of mkdir, it would be helpful if we can test it out once a downstream build with [1] is available. Based on the data available we can decide the course of action for this bz. Putting NEEDINFO on Karan/Ambarish for perf data. regards, Raghavendra.
The needinfo request[s] on this closed bug have been removed as they have been unresolved for 1000 days