You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@couchdb.apache.org by "Filipe Manana (JIRA)" <ji...@apache.org> on 2011/01/05 18:20:45 UTC
[jira] Created: (COUCHDB-1013) branch 1.1.x with worse database
read/write performance
branch 1.1.x with worse database read/write performance
-------------------------------------------------------
Key: COUCHDB-1013
URL: https://issues.apache.org/jira/browse/COUCHDB-1013
Project: CouchDB
Issue Type: Bug
Reporter: Filipe Manana
Fix For: 1.1, 1.2
The issue is described in an e-mail sent to the development mailing list:
http://mail-archives.apache.org/mod_mbox/couchdb-dev/201012.mbox/%3CAANLkTim5uNVkSCVhHuu_1GA1xKVFJFBWE2-Wtb4vVx8g@mail.gmail.com%3E
It also affects trunk.
Would love to have results from other on Linux/ext4, and other OSes/filesystems besides Mac OS X.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Commented: (COUCHDB-1013) branch 1.1.x with worse database
read/write performance
Posted by "Jan Lehnardt (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/COUCHDB-1013?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12994886#comment-12994886 ]
Jan Lehnardt commented on COUCHDB-1013:
---------------------------------------
The following setup:
relaximation runs on a dual 2.13Ghz MacBook Unibody, SSD, 4GB RAM with Mac OS X 10.6.6. It is connected to an Core 2 Duo 3.06Ghz iMac via GigE withMac OS X 10.6.5 on R14B01
Default configuration: http://graphs.mikeal.couchone.com/#/graph/a3dccb9cef53a4910706152d90003649
delayed_commits = false: http://graphs.mikeal.couchone.com/#/graph/a3dccb9cef53a4910706152d900036aa
I don't know if these show anything conclusive. I'm happy to run other configurations if it helps.
> branch 1.1.x with worse database read/write performance
> -------------------------------------------------------
>
> Key: COUCHDB-1013
> URL: https://issues.apache.org/jira/browse/COUCHDB-1013
> Project: CouchDB
> Issue Type: Bug
> Reporter: Filipe Manana
> Fix For: 1.1, 1.2
>
>
> The issue is described in an e-mail sent to the development mailing list:
> http://mail-archives.apache.org/mod_mbox/couchdb-dev/201012.mbox/%3CAANLkTim5uNVkSCVhHuu_1GA1xKVFJFBWE2-Wtb4vVx8g@mail.gmail.com%3E
> It also affects trunk.
> Would love to have results from other on Linux/ext4, and other OSes/filesystems besides Mac OS X.
--
This message is automatically generated by JIRA.
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (COUCHDB-1013) branch 1.1.x with worse database
read/write performance
Posted by "Jan Lehnardt (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/COUCHDB-1013?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13011612#comment-13011612 ]
Jan Lehnardt commented on COUCHDB-1013:
---------------------------------------
I did a few more tests.
First: benchbulk.sh from test/bench/benchbulk.sh
Second: ab
The reasoning, both would allow separate testing of the r/w performance, in the read case (ab), reading a single doc will ensure that we don't measure disk IO, but couch/mochiweb.
I ran benchbulk.sh in two configurations, ab in one configuration, I'm showing the top three out of five runs.
This is on my Unibody MacBook, 2.4Ghz, 4GB RAM, SSD, Erlang R14B (erts-5.8.1) [source] [64-bit] [smp:2:2] [rq:2] [async-threads:4] [hipe] [kernel-poll:true], ab and benchbulk.sh running on the same machine as CouchDB.
Conclusion: No significant differences between cd214b2 and 4b0948d. The results are of limited applicability, but I think they are close enough to support the conclusion.
Results:
benchbulk.sh
BULKSIZE=1000
DOCSIZE=10
INSERTS=10
ROUNDS=10
mochi-pre
real 0m8.553s
user 0m8.296s
sys 0m1.608s
real 0m8.431s
user 0m8.290s
sys 0m1.601s
real 0m8.706s
user 0m8.296s
sys 0m1.603s
mochi-post
real 0m8.412s
user 0m8.287s
sys 0m1.603s
real 0m8.632s
user 0m8.297s
sys 0m1.618s
real 0m8.513s
user 0m8.291s
sys 0m1.608s
--
BULKSIZE=1000
DOCSIZE=100
INSERTS=10
ROUNDS=10
mochi-pre
real 0m11.401s
user 0m12.692s
sys 0m1.796s
real 0m11.998s
user 0m12.707s
sys 0m1.817s
real 0m11.453s
user 0m12.694s
sys 0m1.806s
mochi-post
real 0m11.554s
user 0m12.692s
sys 0m1.793s
real 0m11.616s
user 0m12.690s
sys 0m1.795s
real 0m11.658s
user 0m12.706s
sys 0m1.819s
--
ab -c 25 -n 10000
mochi-pre
Requests per second: 1017.80 [#/sec] (mean)
Time per request: 24.563 [ms] (mean)
Time per request: 0.983 [ms] (mean, across all concurrent requests)
Transfer rate: 307.13 [Kbytes/sec] received
Requests per second: 1084.57 [#/sec] (mean)
Time per request: 23.051 [ms] (mean)
Time per request: 0.922 [ms] (mean, across all concurrent requests)
Transfer rate: 327.28 [Kbytes/sec] received
Requests per second: 1017.11 [#/sec] (mean)
Time per request: 24.580 [ms] (mean)
Time per request: 0.983 [ms] (mean, across all concurrent requests)
Transfer rate: 306.92 [Kbytes/sec] receive
mochi-post
Requests per second: 1051.70 [#/sec] (mean)
Time per request: 23.771 [ms] (mean)
Time per request: 0.951 [ms] (mean, across all concurrent requests)
Transfer rate: 317.36 [Kbytes/sec] received
Requests per second: 1056.78 [#/sec] (mean)
Time per request: 23.657 [ms] (mean)
Time per request: 0.946 [ms] (mean, across all concurrent requests)
Transfer rate: 318.89 [Kbytes/sec] received
Requests per second: 1068.53 [#/sec] (mean)
Time per request: 23.397 [ms] (mean)
Time per request: 0.936 [ms] (mean, across all concurrent requests)
Transfer rate: 322.44 [Kbytes/sec] received
> branch 1.1.x with worse database read/write performance
> -------------------------------------------------------
>
> Key: COUCHDB-1013
> URL: https://issues.apache.org/jira/browse/COUCHDB-1013
> Project: CouchDB
> Issue Type: Bug
> Reporter: Filipe Manana
> Fix For: 1.1, 1.2
>
>
> The issue is described in an e-mail sent to the development mailing list:
> http://mail-archives.apache.org/mod_mbox/couchdb-dev/201012.mbox/%3CAANLkTim5uNVkSCVhHuu_1GA1xKVFJFBWE2-Wtb4vVx8g@mail.gmail.com%3E
> It also affects trunk.
> Would love to have results from other on Linux/ext4, and other OSes/filesystems besides Mac OS X.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (COUCHDB-1013) branch 1.1.x with worse database
read/write performance
Posted by "Filipe Manana (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/COUCHDB-1013?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13011625#comment-13011625 ]
Filipe Manana commented on COUCHDB-1013:
----------------------------------------
Thanks for those comprehensive tests Jan.
On a new Linux installation I compared branch 1.1.x against branch 1.0.x - only set the nodelay option for 1.1.x, as Benoît found recently that under Linux it has a significant positive impact - https://github.com/benoitc/restkit/issues#issue/5
I made a 200 seconds relaximation read and writes test with documents of ~4kb and here are the results:
http://graphs.mikeal.couchone.com/#/graph/698bf36b6c64dbd19aa2bef63404ecbc
These results match yours - 1.1.x has about the same performance as 1.0.x (reads seem slightly better in 1.1.x)
Closing this ticket
> branch 1.1.x with worse database read/write performance
> -------------------------------------------------------
>
> Key: COUCHDB-1013
> URL: https://issues.apache.org/jira/browse/COUCHDB-1013
> Project: CouchDB
> Issue Type: Bug
> Reporter: Filipe Manana
> Fix For: 1.1, 1.2
>
>
> The issue is described in an e-mail sent to the development mailing list:
> http://mail-archives.apache.org/mod_mbox/couchdb-dev/201012.mbox/%3CAANLkTim5uNVkSCVhHuu_1GA1xKVFJFBWE2-Wtb4vVx8g@mail.gmail.com%3E
> It also affects trunk.
> Would love to have results from other on Linux/ext4, and other OSes/filesystems besides Mac OS X.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] Commented: (COUCHDB-1013) branch 1.1.x with worse database
read/write performance
Posted by "Randall Leeds (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/COUCHDB-1013?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12979227#action_12979227 ]
Randall Leeds commented on COUCHDB-1013:
----------------------------------------
Filipe: I cannot confirm your results. In my testing it seems write performance is slightly better while read performance is slightly worse.
First test, default settings
1.0.x@2a1f5e8b4a8bf80c632d44b6c81b0d9dfbd55174 1.1.x@38b836a044900545834f6a6a5126d09b244379cf
http://graphs.mikeal.couchone.com/#/graph/8bf31813eef7c0b7e37d1ea25903acfe
Second test, cherry pick COUCHDB-1008 onto 1.1.x
1.0.x@2a1f5e8b4a8bf80c632d44b6c81b0d9dfbd55174
1.1.x@ac4c65a4712407bb64edaca5739c0ae4acc49c02
1.1.x with server_options= [{acceptor_pool_size, 1}, {nodelay, true}]
http://graphs.mikeal.couchone.com/#/graph/8bf31813eef7c0b7e37d1ea25903c8c6
Third test, same versions as the second
1.0.x@2a1f5e8b4a8bf80c632d44b6c81b0d9dfbd55174
1.1.x@ac4c65a4712407bb64edaca5739c0ae4acc49c02
server_options= [{acceptor_pool_size, 16}, {nodelay, true}]
1.1.x with server_options= [{acceptor_pool_size, 16}, {nodelay, true}]
It looks like {nodelay, true} gives me slightly worse performance.
I got slightly better read performance, but worse write performance, with only 1 acceptor process if you compare the second and third runs, but it may be insignificant.
My read results are strange; every run shows two plateaus for read throughput with the better parts on par with 1.0.x but most of the samples far worse.
This was all done on my laptop, Ubuntu 10.10, Erlang R13B03. I can possibly repeat the tests on Monday on server class hardware. These results don't show strictly worse performance for 1.1.x and IMO is not necessarily a blocker.
> branch 1.1.x with worse database read/write performance
> -------------------------------------------------------
>
> Key: COUCHDB-1013
> URL: https://issues.apache.org/jira/browse/COUCHDB-1013
> Project: CouchDB
> Issue Type: Bug
> Reporter: Filipe Manana
> Fix For: 1.1, 1.2
>
>
> The issue is described in an e-mail sent to the development mailing list:
> http://mail-archives.apache.org/mod_mbox/couchdb-dev/201012.mbox/%3CAANLkTim5uNVkSCVhHuu_1GA1xKVFJFBWE2-Wtb4vVx8g@mail.gmail.com%3E
> It also affects trunk.
> Would love to have results from other on Linux/ext4, and other OSes/filesystems besides Mac OS X.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Commented: (COUCHDB-1013) branch 1.1.x with worse database
read/write performance
Posted by "Filipe Manana (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/COUCHDB-1013?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12979285#action_12979285 ]
Filipe Manana commented on COUCHDB-1013:
----------------------------------------
Randall thanks for trying this out too.
One thing I noticed is that you ran the tests with delayed_commits set to true. Can you try with it set to false? (this is how I ran them)
My tests were done on Ubuntu 10.10 as well, but with OTP R14B01. I didn't see any significant differences here when changing the acceptor pool size and/or nodelay.
Btw, if you compare 1.1.x against 1.0.x, what do you see?
thanks again
> branch 1.1.x with worse database read/write performance
> -------------------------------------------------------
>
> Key: COUCHDB-1013
> URL: https://issues.apache.org/jira/browse/COUCHDB-1013
> Project: CouchDB
> Issue Type: Bug
> Reporter: Filipe Manana
> Fix For: 1.1, 1.2
>
>
> The issue is described in an e-mail sent to the development mailing list:
> http://mail-archives.apache.org/mod_mbox/couchdb-dev/201012.mbox/%3CAANLkTim5uNVkSCVhHuu_1GA1xKVFJFBWE2-Wtb4vVx8g@mail.gmail.com%3E
> It also affects trunk.
> Would love to have results from other on Linux/ext4, and other OSes/filesystems besides Mac OS X.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] [Closed] (COUCHDB-1013) branch 1.1.x with worse database
read/write performance
Posted by "Filipe Manana (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/COUCHDB-1013?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Filipe Manana closed COUCHDB-1013.
----------------------------------
Resolution: Not A Problem
> branch 1.1.x with worse database read/write performance
> -------------------------------------------------------
>
> Key: COUCHDB-1013
> URL: https://issues.apache.org/jira/browse/COUCHDB-1013
> Project: CouchDB
> Issue Type: Bug
> Reporter: Filipe Manana
> Fix For: 1.1, 1.2
>
>
> The issue is described in an e-mail sent to the development mailing list:
> http://mail-archives.apache.org/mod_mbox/couchdb-dev/201012.mbox/%3CAANLkTim5uNVkSCVhHuu_1GA1xKVFJFBWE2-Wtb4vVx8g@mail.gmail.com%3E
> It also affects trunk.
> Would love to have results from other on Linux/ext4, and other OSes/filesystems besides Mac OS X.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira