[
https://issues.apache.org/jira/browse/CASSANDRA-1470?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12933188#action_12933188
]
T Jake Luciani edited comment on CASSANDRA-1470 at 11/17/10 5:29 PM:
---------------------------------------------------------------------
Couple more fixes:
1) Tweaked directio buffer size to be 16k. Gets much better read performance.
2) Another fix for non-jna users
I've done some substantial testing of this patch and I have to say I see no
effect. I think the reason is commit log and SSTable writes are entering the
page cache as well. I think the next approach would be to make these use
DirectIO as well. Back to you Pavel :)
My benchmarking looks like this:
{code}
#Write a lot of data to CF
python stress.py -S 8192 -n 100000 -k
#triggers minor compaction
#wait till complete....
#read from CF till file cache is primed
python stress.py -n 100000 -k -o read
total,interval_op_rate,interval_key_rate,avg_latency,elapsed_time
1010,101,101,0.5461584162,11
2465,145,145,0.384013745219,22
[cropped]
#primed read
python stress.py -n 100000 -k -o read
total,interval_op_rate,interval_key_rate,avg_latency,elapsed_time
26962,2696,2696,0.0219141756796,11
59567,3260,3260,0.0170967640897,22
97089,3752,3752,0.0146972443858,33
100000,291,290,0.00880796890495,34
#write to another column family
python stress.py -S 8192 -n 100000 -k -y super
#triggers minor compaction of second CF.. wait till complete
#read from first CF again (slowwww)
python stress.py -n 100000 -k -o read
total,interval_op_rate,interval_key_rate,avg_latency,elapsed_time
1010,101,101,0.5461584162,11
2465,145,145,0.384013745219,22
3915,145,145,0.385855172749,33
5683,176,176,0.307302036167,44
7591,190,190,0.291640462241,55
9575,198,198,0.281985886275,66
[cropped]
{code}
was (Author: tjake):
Couple more fixes:
1) Tweaked directio buffer size to be 16k. Gets much better read performance.
2) Another fix for non-jna users
I've done some substantial testing of this patch and I have to say I see no
effect. I think the reason is commit log and SSTable writes are entering the
page cache as well. I think the next approach would be to make these use
DirectIO as well. Back to you Pavel :)
My benchmarking looks like this:
{code}
#Write a lot of data to 2 column families
python stress.py -S 8192 -n 100000 -k
python stress.py -S 8192 -n 100000 -k -y super
#read it till file cache is primed
python stress.py -n 100000 -k -o read
total,interval_op_rate,interval_key_rate,avg_latency,elapsed_time
1010,101,101,0.5461584162,11
2465,145,145,0.384013745219,22
[cropped]
#primed read
python stress.py -n 100000 -k -o read
total,interval_op_rate,interval_key_rate,avg_latency,elapsed_time
26962,2696,2696,0.0219141756796,11
59567,3260,3260,0.0170967640897,22
97089,3752,3752,0.0146972443858,33
100000,291,290,0.00880796890495,34
#force compaction
nodetool -h localhost compact Keyspace1
#read (slowwww)
python stress.py -n 100000 -k -o read
total,interval_op_rate,interval_key_rate,avg_latency,elapsed_time
1010,101,101,0.5461584162,11
2465,145,145,0.384013745219,22
3915,145,145,0.385855172749,33
5683,176,176,0.307302036167,44
7591,190,190,0.291640462241,55
9575,198,198,0.281985886275,66
[cropped]
{code}
> use direct io for compaction
> ----------------------------
>
> Key: CASSANDRA-1470
> URL: https://issues.apache.org/jira/browse/CASSANDRA-1470
> Project: Cassandra
> Issue Type: Improvement
> Components: Core
> Reporter: Jonathan Ellis
> Assignee: Pavel Yaskevich
> Fix For: 0.6.9
>
> Attachments: 1470-v2.txt, 1470.txt, CASSANDRA-1470-for-0.6.patch,
> CASSANDRA-1470-v10-for-0.7.patch, CASSANDRA-1470-v11-for-0.7.patch,
> CASSANDRA-1470-v2.patch,
> CASSANDRA-1470-v3-0.7-with-LastErrorException-support.patch,
> CASSANDRA-1470-v4-for-0.7.patch, CASSANDRA-1470-v5-for-0.7.patch,
> CASSANDRA-1470-v6-for-0.7.patch, CASSANDRA-1470-v7-for-0.7.patch,
> CASSANDRA-1470-v8-for-0.7.patch, CASSANDRA-1470-v9-for-0.7.patch,
> CASSANDRA-1470.patch,
> use.DirectIORandomAccessFile.for.commitlog.against.1022235.patch
>
>
> When compaction scans through a group of sstables, it forces the data in the
> os buffer cache being used for hot reads, which can have a dramatic negative
> effect on performance.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
