You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@cassandra.apache.org by "Wei Deng (JIRA)" <ji...@apache.org> on 2016/09/05 03:24:21 UTC

[jira] [Created] (CASSANDRA-12610) Introduce a mixed write + compact mode for compaction-stress to simulate more realistic workload

Wei Deng created CASSANDRA-12610:
------------------------------------

             Summary: Introduce a mixed write + compact mode for compaction-stress to simulate more realistic workload
                 Key: CASSANDRA-12610
                 URL: https://issues.apache.org/jira/browse/CASSANDRA-12610
             Project: Cassandra
          Issue Type: Improvement
          Components: Compaction
            Reporter: Wei Deng


The new offline stress utility {{compaction-stress}} was introduced in CASSANDRA-11844 which greatly simplified the amount of work needed to perform a compaction experiment. However, it currently only provides two modes {{write}} and {{compact}}, which means the user will always have to use {{write}} to generate a bunch of SSTables first, and then use {{compact}} to compact them using desired compaction strategy. This is not as close to the real workload as possible where new SSTable writing and compaction can often happen simultaneously which puts a different strain on CPU and IO than the current test pattern.

With the introduction of the mixed (write / compact) mode proposed in this JIRA, we should at least allow parameters to specify how big each newly-written SSTable is (already supported by {{buffer-size-mb}}), and how frequently the write happens (this is the most straightforward way to simulate a memtable flush and might over-simplify the flush behavior, but should be easier to implement; the more sophisticated SSTable write behavior probably can be done in a follow-up JIRA), and allow specifying how many threads allocated to {{write}} and {{compact}} respectively.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)