[ https://issues.apache.org/jira/browse/CASSANDRA-18104?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17646384#comment-17646384 ]
Sreedhar J commented on CASSANDRA-18104: ---------------------------------------- !screenshot-1.png! Please see the attached screenshot. I did a bit of analysis on the output when you have TRACING ON. I warmed up my system by running the query 100 times. Then collected the output from 30 iterations of the query. I did this for both versions. I calculated the time between each 'activity' in the tracing to get a sense for which activities were taking longest. I then calculated the average, min, max time spent on each across all 30 iterations I compared them between 3.11.14 and 4.0.7 to see if I could spot a consistent place where Cassandra was spending more time on 4.0.7. The numbers vary greatly each time I run the query, but here's a table that shows the data: If you look at #10, you can see that the Average is 617 microseconds higher, the minimum value is 435 microseconds higher and the maximums are 4170 higher. > Major Performance degradation of Casandara 4.0.7 against Casandra 3.11.14 > ---------------------------------------------------------------------------- > > Key: CASSANDRA-18104 > URL: https://issues.apache.org/jira/browse/CASSANDRA-18104 > Project: Cassandra > Issue Type: Bug > Reporter: Sreedhar J > Priority: Urgent > Attachments: 3.11.14.txt, 4.0.7.txt, mailboxes_snapshot.zip, > query30.txt, screenshot-1.png > > > Our application uses Casandra 3.11.x and has lot of security vulnerabilities > which are addressed in 4.0.x. So we have upgraded the Casandra to 4.0.7 > and our performance tests have shown aorund 20% degradation compare to > 3.11.x > We are now able to reproduce the same performance degradation using the > standalone queries. Here are the steps. > 1. Expand Cassandra 3.11.14 tarball and 4.0.7 tarball to different folders > 2. Import the attached data from the snapshot (mailboxes_snapshot.zip) into > each Cassandra instance, see schema.cql for CQL for creating the required > table and indexes before import > 3. With CQLSH run the following query several times with TRACING ON and > PAGING OFF against both versions of Cassandra: select * from > mailbox.mailboxes where mbx_id= 6c57da2e-7ddd-4984-be62-105415e6b48a; > 4. Compare results > IWe ran the target query 30 times. Here's the average times to run the query: > 3.11.14 - 19400.77 > 4.0.7 - 34906.03 -- This message was sent by Atlassian Jira (v8.20.10#820010) --------------------------------------------------------------------- To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org