Actions
Bug #6372
closedrados bench reports bad statistics when read --block-size > write --block-size
Status:
Resolved
Priority:
Normal
Assignee:
-
Category:
-
Target version:
-
% Done:
0%
Source:
other
Tags:
Backport:
Regression:
No
Severity:
3 - minor
Reviewed:
Affected Versions:
ceph-qa-suite:
Pull request ID:
Crash signature (v1):
Crash signature (v2):
Description
The current MB/s and average MB/s statistics for the rados read benchmark are off when the specified block size for the read is greater than the write block size. The statistics seem to be off by a factor of "seq --block-size" divided by "write --block-size".
Since the object size is established by the write benchmark, the "read size" that appears in the summary is correct, and the calculated bandwidth is therefore correct.
I've attached an example that illustrates this.
Files
Updated by Dan Mick over 10 years ago
I suspect the read is issued for > object size, but then just completes early, but the
stats tracking still uses the request size as "the amount of I/O done". I suggested
the experiment of halving the read size, but this unfortunately segfaults (see #6371).
Updated by Loïc Dachary almost 9 years ago
- Status changed from New to Duplicate
- Regression set to No
Updated by Loïc Dachary almost 9 years ago
- Status changed from 12 to Resolved
$ rados bench -p rbd 60 seq -t 32 --block-size=16777216 -b|--block_size option can be used only with `write' bench test
Actions