[
https://issues.apache.org/jira/browse/CASSANDRA-9738?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14739436#comment-14739436
]
Ariel Weisberg commented on CASSANDRA-9738:
-------------------------------------------
[~slebresne]
For the OHC key cache we are looking at rolling back the core changes for
#10232 because they get in the way of accessing the data off heap when we want
to random access fields. Also the RIE is sampling so the value of slightly
shrinking the entries seems low to me. There is a time/space tradeoff to make
since we have to binary search the indexed RIE.
Some of the other optimizations don't work like dropping fields and
recalculating them because we don't want to process the index to read it just
copy the bytes. I think this also prevented the optimization where you rebased
the offset at 64k because that also requires a pass through the data to parse.
I asked snazy to benchmark with large partitions with and without vints, but I
think it's splitting hairs either way.
It's a bad situation in general because when we cache/uncache these things we
copy the entire thing which is yucky. It will be faster this way because at
least it is a straight memory copy without an object graph or parsing.
My pony for 3.0 would be to 64-bit map the entire index file, then get 32-bit
ByteBuffer slices out of it and not use the key cache entirely, but there is a
time constraint.
I think just mapping it naively might get us 80 in the 80/20 of CASSANDRA-9754.
Benedict also mentioned CASSANDRA-8931 which sounds like good next release
fodder for making the index smaller.
> Migrate key-cache to be fully off-heap
> --------------------------------------
>
> Key: CASSANDRA-9738
> URL: https://issues.apache.org/jira/browse/CASSANDRA-9738
> Project: Cassandra
> Issue Type: Sub-task
> Reporter: Robert Stupp
> Assignee: Robert Stupp
> Fix For: 3.0.0 rc1
>
>
> Key cache still uses a concurrent map on-heap. This could go to off-heap and
> feels doable now after CASSANDRA-8099.
> Evaluation should be done in advance based on a POC to prove that pure
> off-heap counter cache buys a performance and/or gc-pressure improvement.
> In theory, elimination of on-heap management of the map should buy us some
> benefit.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
