0

when I try to use cosbench to test my cluster's performance, I find that performance of object storage swift fluctuates every 30 seconds. An experiment has been done as follows:

1 proxy server and 5 storage server( 3 for object server and the other 2 for account/container server), each server have 8*1T HDD, 64GB memory and 12 CPU cores. Just the service object server is launched. when only one client puts small objects to the cluster, the phenomenon occurs. I find that many disk requests arrive and all the disks latency rise suddenly every 30 seconds. Anyone can tell me why this happens.

2 Answers2

0

Default run_pause argument is set to 30 seconds. In every 30 seconds the object-replicator, container-replicator and account-replicator start to work.

small_ticket
  • 1,910
  • 5
  • 22
  • 30
  • Finally I found out the root cause. The xfs file system will sync files every 30 seconds, that caused the performance degradation. – trueman1990 Nov 19 '14 at 13:00
0

Finally I found out the root cause. The xfs file system will sync files every 30 seconds, that caused the performance degradation.