Details
-
Bug
-
Resolution: Fixed
-
Major
-
Cheshire-Cat
-
Triaged
-
1
-
Unknown
Description
We ran the Magma YCSB DP test on the nvme cluster. Both runs are using the same build (1006.5.1-1128) and the same test configure.
DP: http://perf.jenkins.couchbase.com/job/rhea-ycsb/856/
Logs:
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-rhea-ycsb-856/172.23.97.25.zip
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-rhea-ycsb-856/172.23.97.26.zip
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-rhea-ycsb-856/172.23.97.29.zip
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-rhea-ycsb-856/172.23.97.30.zip
nvme: http://perf.jenkins.couchbase.com/job/magma-nvme-ycsb/185/
Logs:
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-magma-nvme-ycsb-185/172.23.100.135.zip
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-magma-nvme-ycsb-185/172.23.100.136.zip
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-magma-nvme-ycsb-185/172.23.100.137.zip
https://s3-us-west-2.amazonaws.com/perf-artifacts/jenkins-magma-nvme-ycsb-185/172.23.100.138.zip
Compared to the DP run, the new run on the nvme cluster has lower performance. The nvme cluster has more CPU cores and higher CPU clock so it should have better performance.
When we compare these two runs, we see that the nvme run has lower resident ratio and much higher dcp_replica_items_remaining.
Attachments
Issue Links
- relates to
-
MB-40510 number of vb_replica_curr_items isn't equal to number of curr_items in Magma YCSB runs
- Closed