[
https://issues.apache.org/jira/browse/CASSANDRA-6890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13948366#comment-13948366
]
Joshua McKenzie commented on CASSANDRA-6890:
--------------------------------------------
See CASSANDRA-4050 for further details - ran into some performance issues with
conversion to nio.2 and tracked it down to me combining a bypass on mmap'ed
data on that patch. Removed that portion of things on that patch and will work
further on getting to the bottom of that here. Currently a simple switch from
a memory mapped to non-memory-mapped file in SegmentedFile's getBuilder is
showing a 15-40% drop in throughput depending on work-load characteristics.
Some more detailed yammer timers on small subsets of the code (a single
FileChannel seek, a single byte read operation) show that #'s are comparable up
to somewhere between 50th and 75th percentile and then they balloon upwards
from there.
I'll run jstat against a jvm in both scenarios and see what it shows.
> Remove mmap'ed i/o from read path
> ---------------------------------
>
> Key: CASSANDRA-6890
> URL: https://issues.apache.org/jira/browse/CASSANDRA-6890
> Project: Cassandra
> Issue Type: Improvement
> Reporter: Joshua McKenzie
> Assignee: Joshua McKenzie
> Priority: Minor
> Fix For: 3.0
>
> Attachments: mmap_perf.txt
>
>
> Since we actively unmap unreferenced SSTR's and also copy data out of those
> readers on the read path, the current memory mapped i/o is a lot of
> complexity for very little payoff. Clean out the mmapp'ed i/o on the read
> path.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
