Description of problem: While running tests with the upstream master (HEAD at commit dfa255ae7f2dab4fb3d84c67a0452c5b32455877), I noticed that enabling parallel-readdir seems to increase the time taken for a directory listing: Numbers from a pure distribute 3 brick volume: Volume Name: pvol Type: Distribute Volume ID: c39c8c16-82d3-4b0b-8050-9c3d22c800ea Status: Started Snapshot Count: 0 Number of Bricks: 3 Transport-type: tcp Bricks: Brick1: server1:/mnt/bricks/fsgbench0002/brick-0 Brick2: server1:/mnt/bricks/fsgbench0003/brick-0 Brick3: server1:/mnt/bricks/fsgbench0004/brick-0 Options Reconfigured: transport.address-family: inet nfs.disable: on The volume was mounted on /mnt/nithya and I created 10K directories and 10K files in the volume root: With readdir-ahead enabled: ---------------------------- [root@server2 nithya]# time ll |wc -l 20001 real 0m11.434s user 0m0.116s sys 0m0.241s [root@server2 nithya]# time ll |wc -l 20001 real 0m6.825s user 0m0.111s sys 0m0.265s With readdir-ahead and parallel-readdir enabled: ------------------------------------------------ [root@server2 nithya]# time ll |wc -l 20001 real 0m15.609s user 0m0.148s sys 0m0.379s [root@server2 nithya]# time ll |wc -l 20001 real 0m9.930s user 0m0.107s sys 0m0.295s
This bug is moved to https://github.com/gluster/glusterfs/issues/842, and will be tracked there from now on. Visit GitHub issues URL for further details