[prev in list] [next in list] [prev in thread] [next in thread] 

List:       ceph-devel
Subject:    CephFS client performance issue
From:       Marvin Zhang <fanzier () gmail ! com>
Date:       2018-12-19 4:32:17
Message-ID: CA+5eE-Tb+5PgBiqGDx83Wk3xNMtw90_HohYSvBtoQxvjuctX9w () mail ! gmail ! com
[Download RAW message or body]

Hi all,
Here is my test result for single MDS with single fuse client. I read
100,000  different files which are all size of 64K.  Fuse client is on
the same machine of MDS.  The MDS and fuse client log level are set to
0. The fuse client latency is about 20 times of MDS. Is it as
expected?
I found that each request on client will try to accquire a lock, even
though it will release when IO happend. Is it caused the latency?

ceph daemon <mds.asok> perf dump | jq '.mds'
  "request": 99990,
  "reply": 99990,
  "reply_latency": {
    "avgcount": 99990,
    "sum": 9.653804187,
    "avgtime": 9.6547e-05
  }

ceph daemon <client.asok> perf dump | jq '.mds'
  "reply": {
    "avgcount": 99990,
    "sum": 181.812120526,
    "avgtime": 0.001818303
  }
Marvin
[prev in list] [next in list] [prev in thread] [next in thread]