[
https://issues.apache.org/jira/browse/ACCUMULO-1755?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15184994#comment-15184994
]
Dave Marion commented on ACCUMULO-1755:
---------------------------------------
I wrote a new test that sends 1M mutations total using N threads with a
BatchWriter buffer of different sizes. The test is run twice and the time
discarded to account for JVM startup. Then the test is run 10 times and the
average (in seconds) is reported for total time and time to add mutations.
First, I added some code to the TSBW to determine that with my test data I was
sending the following number of batches using 1, 10, and 100MB buffers:
||BatchWriter Max Memory Size || Flushes To Accumulo
| 1MB | 515 |
| 10MB | 52 |
| 100MB | 6 |
Here are the results of the test:
h2. master branch
Using the patch 1755-perf-test.patch
Total Time
||Threads|| 1MB || 10MB || 100MB ||
| 1 | 3.121 | 2.818 | 3.158 |
| 2 | 3.102 | 2.414 | 2.950 |
| 4 | 3.367 | 2.573 | 3.114 |
| 8 | 3.422 | 2.569 | 3.140 |
| 16 | 3.590 | 2.741 | 3.332 |
Add Mutation Time
||Threads|| 1MB || 10MB || 100MB ||
| 1 | 3.114 | 2.733 | 2.498 |
| 2 | 3.088 | 2.350 | 2.371 |
| 4 | 3.360 | 2.506 | 2.472 |
| 8 | 3.414 | 2.516 | 2.509 |
| 16 | 3.582 | 2.692 | 2.696 |
h2. master branch with modifications to remove sync on addMutation()
I successfully modified the TSBW to remove the synchronization modifier from
the addMutation method. The multi-threaded binning test passes so I have some
confidence that the data is correct. Use patch 1755-nosync-perf-test.patch
Total Time
||Threads|| 1MB || 10MB || 100MB ||
| 1 | 3.080 | 2.766 | 3.255 |
| 2 | 2.972 | 2.420 | 3.137 |
| 4 | 3.162 | 2.492 | 3.190 |
| 8 | 3.100 | 2.658 | 3.623 |
| 16 | 3.393 | 2.898 | 3.743 |
Add Mutation Time
||Threads|| 1MB || 10MB || 100MB ||
| 1 | 3.072 | 2.653 | 2.517 |
| 2 | 2.965 | 2.371 | 2.527 |
| 4 | 3.155 | 2.441 | 2.589 |
| 8 | 3.092 | 2.602 | 2.961 |
| 16 | 3.385 | 2.839 | 2.891 |
I think the results are inconclusive. The tests run with MAC on localhost, so
this is likely a best case scenario. I'd be interested to see this re-run on a
real cluster.
> BatchWriter blocks all addMutation calls while binning mutations
> ----------------------------------------------------------------
>
> Key: ACCUMULO-1755
> URL: https://issues.apache.org/jira/browse/ACCUMULO-1755
> Project: Accumulo
> Issue Type: Improvement
> Components: client
> Reporter: Adam Fuchs
> Assignee: Dave Marion
> Fix For: 1.6.6, 1.7.2, 1.8.0
>
> Attachments: ACCUMULO-1755.patch
>
> Time Spent: 2h
> Remaining Estimate: 0h
>
> Through code inspection, we found that the BatchWriter bins mutations inside
> of a synchronized block that covers calls to addMutation. Binning potentially
> involves lookups of tablet metadata and processes a fair amount of
> information. We will get better parallelism if we can either unlock the lock
> while binning, dedicate another thread to do the binning, or use one of the
> send threads to do the binning.
> This has not been verified empirically yet, so there is not yet any profiling
> info to indicate the level of improvement that we should expect. Profiling
> and repeatable demonstration of this performance bottleneck should be the
> first step on this ticket.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
