[
https://issues.apache.org/jira/browse/CASSANDRA-7409?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14277372#comment-14277372
]
Carl Yeksigian commented on CASSANDRA-7409:
-------------------------------------------
I've pushed up an updated branch at:
https://github.com/carlyeks/cassandra/tree/overlapping-better-selection
The compaction selection has been updated so that it performs much better now.
The biggest issue was selecting a single sstable for the overlapping
compactions, instead of the that one plus the overlapping ones. It performs
much better now.
|| || Operation Time || Compaction Time ||
| MOLO=0 | 3:52:46 | 0:21:04 |
| MOLO=2 | 3:45:52 | 0:37:50 |
| MOLO=5 | 3:42:59 | 0:03:17 |
| LCS w/ STCS | 3:48:14 | 0:50:24 |
| LCS w/o STCS | 3:50:38 | 1:05:02 |
The performance on spinning disk is also improved by allowing overlapping; here
are the results of a read operation after running a large mixed read/write
workload:
http://cstar.datastax.com/graph?stats=e113f706-9b54-11e4-9f2c-42010af0688f&metric=op_rate&operation=2_user&smoothing=1&show_aggregates=true&xmin=0&xmax=121.88&ymin=0&ymax=113984.2
> Allow multiple overlapping sstables in L1
> -----------------------------------------
>
> Key: CASSANDRA-7409
> URL: https://issues.apache.org/jira/browse/CASSANDRA-7409
> Project: Cassandra
> Issue Type: Improvement
> Reporter: Carl Yeksigian
> Assignee: Carl Yeksigian
> Labels: compaction
> Fix For: 3.0
>
>
> Currently, when a normal L0 compaction takes place (not STCS), we take up to
> MAX_COMPACTING_L0 L0 sstables and all of the overlapping L1 sstables and
> compact them together. If we didn't have to deal with the overlapping L1
> tables, we could compact a higher number of L0 sstables together into a set
> of non-overlapping L1 sstables.
> This could be done by delaying the invariant that L1 has no overlapping
> sstables. Going from L1 to L2, we would be compacting fewer sstables together
> which overlap.
> When reading, we will not have the same one sstable per level (except L0)
> guarantee, but this can be bounded (once we have too many sets of sstables,
> either compact them back into the same level, or compact them up to the next
> level).
> This could be generalized to allow any level to be the maximum for this
> overlapping strategy.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
