I get better throughput for some workloads after disabling journal compression. At least one other JIRA was already open for perf problems with this –
My request is for a configuration option to disable it. If that is not feasible then the current code should be more clever about hiding latency from compression, like starting to compress earlier rather than in Journal::journal.
From other testing I see that MongoDB writes much more to the journal than InnoDB writes to its redo log so making MongoDB less verbose will also help. See http://smalldatum.blogspot.com/2014/03/redo-logs-in-mongodb-and-innodb.html
- related to
SERVER-9802 Single-threaded journal compression becomes a bottleneck when using "durable" writes