[
https://issues.apache.org/jira/browse/CASSANDRA-7282?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14105022#comment-14105022
]
Benedict commented on CASSANDRA-7282:
-------------------------------------
Ryan provided me with logs, and we can see a flush trigger exactly as the 0.5
mct write test finishes, explaining the dip. However the temporary dip below
normal is a bit odd and I'd like to explain it better - best guess is key cache
being populated; there is a slower ramp up for the normal run as well, here,
which would be consistent with it (since it flushed earlier). We could try a
run with key cache disabled, to confirm.
I'd also like to be able to explain the different GC characteristics - we GC
less often with the new code, but have one _very_ long ParNew pause, fairly
consistently. Currently I don't have a great explanation for this, except
possibly the very large array allocations we need to do. Possibly switching to
an array-of-arrays would be superior, keeping each of the arrays preferably <
1M or so
For the mct 0.99 run, what's most surprising is reads are _not_ particularly
faster, which is not what I found running locally, and also conflicts with the
non-trivial increase in write performance. Back to the drawing board to try and
figure out exactly what's happening.
> Faster Memtable map
> -------------------
>
> Key: CASSANDRA-7282
> URL: https://issues.apache.org/jira/browse/CASSANDRA-7282
> Project: Cassandra
> Issue Type: Improvement
> Components: Core
> Reporter: Benedict
> Assignee: Benedict
> Labels: performance
> Fix For: 3.0
>
>
> Currently we maintain a ConcurrentSkipLastMap of DecoratedKey -> Partition in
> our memtables. Maintaining this is an O(lg(n)) operation; since the vast
> majority of users use a hash partitioner, it occurs to me we could maintain a
> hybrid ordered list / hash map. The list would impose the normal order on the
> collection, but a hash index would live alongside as part of the same data
> structure, simply mapping into the list and permitting O(1) lookups and
> inserts.
> I've chosen to implement this initial version as a linked-list node per item,
> but we can optimise this in future by storing fatter nodes that permit a
> cache-line's worth of hashes to be checked at once, further reducing the
> constant factor costs for lookups.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
