[ https://issues.apache.org/jira/browse/CASSANDRA-18027?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17632708#comment-17632708 ]
Josh McKenzie edited comment on CASSANDRA-18027 at 11/12/22 5:18 PM: --------------------------------------------------------------------- bq. Benchmarks show that G1 and Cassandra work best when objects are promoted immediately Curious - were those in-memory max throughput, or fixed velocity latency, or spilling over to disk dealing with large workloads? And to be clear, +1 on the general sentiment of what you're doing here and this ticket. I'm just noodling on how we benchmark things as a project historically and whether there'd be value in clarifying when and how we benchmark different things; well outside the scope of this ticket (as is my apparent habit - sorry =/) edit: at least for the TLP benchmarks I can actually read a bit and answer my own questions: {quote} Benchmarks were done using 8 threads running with rate limiting and 80% writes/20% reads. tlp-stress uses asynchronous queries extensively, which can easily overwhelm Cassandra nodes with a limited number of stress threads. The load tests were conducted with each thread sending 50 concurrent queries at a time. The keyspace was created with a replication factor of 3 and all queries were executed at consistency level LOCAL_ONE. All workloads ran for 30 minutes, loading approximately 5 to 16 GB of data per node and allowing a reasonable compaction load. {quote} was (Author: jmckenzie): bq. Benchmarks show that G1 and Cassandra work best when objects are promoted immediately Curious - were those in-memory max throughput, or fixed velocity latency, or spilling over to disk dealing with large workloads? And to be clear, +1 on the general sentiment of what you're doing here and this ticket. I'm just noodling on how we benchmark things as a project historically and whether there'd be value in clarifying when and how we benchmark different things; well outside the scope of this ticket (as is my apparent habit - sorry =/) > Use G1GC as default > ------------------- > > Key: CASSANDRA-18027 > URL: https://issues.apache.org/jira/browse/CASSANDRA-18027 > Project: Cassandra > Issue Type: Task > Components: Local/Config > Reporter: Michael Semb Wever > Assignee: Michael Semb Wever > Priority: Normal > Fix For: 4.x > > > G1GC is well battle tested now, and the recommended configuration for most > users. CMS can work well on smaller heaps but requires more tuning, initially > and over time. G1GC just works. CMS was deprecated in JDK 9. > Patch at > https://github.com/apache/cassandra/compare/trunk...thelastpickle:cassandra:mck/7486/trunk -- 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