[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15602781#comment-15602781
]
stack commented on HBASE-16890:
-------------------------------
Ran the test with 10 columns and a single node DFS cluster. I see the async 5%
faster:
Classic:
{code}
8031674 Performance counter stats for './hbase/bin/hbase --config
/home/stack/conf_hbase org.apache.hadoop.hbase.wal.WALPerformanceEvaluation
-threads 100 -ite rations 250000 -keySize 50 -valueSize 100 -qualifiers
10':
8031675
8031676 5441780.264390 task-clock (msec) # 9.846 CPUs utilized
8031677 40,118,809 context-switches # 0.007 M/sec
8031678 6,179,887 cpu-migrations # 0.001 M/sec
8031679 23,019,942 page-faults # 0.004 M/sec
8031680 10,534,132,474,719 cycles # 1.936 GHz
8031681 <not supported> stalled-cycles-frontend
8031682 <not supported> stalled-cycles-backend
8031683 3,198,340,897,844 instructions # 0.30 insns per cycle
8031684 546,248,857,283 branches # 100.381 M/sec
8031685 8,545,082,370 branch-misses # 1.56% of all branches
8031686
8031687 552.674957713 seconds time elapsed
{code}
Async:
{code}
14570 Performance counter stats for './hbase/bin/hbase --config
/home/stack/conf_hbase org.apache.hadoop.hbase.wal.WALPerformanceEvaluation
-threads 100 -itera tions 250000 -keySize 50 -valueSize 100 -qualifiers
10':
14571
14572 4993770.973851 task-clock (msec) # 9.525 CPUs utilized
14573 34,198,406 context-switches # 0.007 M/sec
14574 6,591,365 cpu-migrations # 0.001 M/sec
14575 14,969,734 page-faults # 0.003 M/sec
14576 10,083,281,472,992 cycles # 2.019 GHz
14577 <not supported> stalled-cycles-frontend
14578 <not supported> stalled-cycles-backend
14579 2,934,520,140,301 instructions # 0.29 insns per cycle
14580 500,829,859,305 branches # 100.291 M/sec
14581 7,413,365,504 branch-misses # 1.48% of all branches
14582
14583 524.262212066 seconds time elapsed
{code}
These two WAL implementations are hard to compare since they are so different
but this compare is important since it removes fanout from the compare; it
should be possible to make asyncwal better than dfsclient in this case.
> Analyze the performance of AsyncWAL and fix the same
> ----------------------------------------------------
>
> Key: HBASE-16890
> URL: https://issues.apache.org/jira/browse/HBASE-16890
> Project: HBase
> Issue Type: Sub-task
> Components: wal
> Affects Versions: 2.0.0
> Reporter: ramkrishna.s.vasudevan
> Assignee: ramkrishna.s.vasudevan
> Fix For: 2.0.0
>
> Attachments: contention.png, contention_defaultWAL.png
>
>
> Tests reveal that AsyncWAL under load in single node cluster performs slower
> than the Default WAL. This task is to analyze and see if we could fix it.
> See some discussions in the tail of JIRA HBASE-15536.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
