[
https://issues.apache.org/jira/browse/HBASE-22564?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16862766#comment-16862766
]
Duo Zhang commented on HBASE-22564:
-----------------------------------
And for scan, the new implementation is way better.
{noformat}
./bin/hbase pe --nomapred --rows=1000000 scan 10 &>test_scan.log < /dev/null &
{noformat}
master
{noformat}
2019-06-13 13:58:27,159 INFO [main] hbase.PerformanceEvaluation: [ScanTest]
Summary of timings (ms): [20897, 20982, 20569, 21168, 20766, 20826, 20231,
21049, 21041, 14057]
2019-06-13 13:58:27,260 INFO [main] hbase.PerformanceEvaluation: [ScanTest
duration ] Min: 14057ms Max: 21168ms Avg: 20158ms
2019-06-13 13:58:27,261 INFO [main] hbase.PerformanceEvaluation: [ Avg latency
(us)] 20
2019-06-13 13:58:27,261 INFO [main] hbase.PerformanceEvaluation: [ Avg
TPS/QPS] 503147 row per second
{noformat}
HBASE-21512
{noformat}
2019-06-13 14:00:31,062 INFO [main] hbase.PerformanceEvaluation: [ScanTest]
Summary of timings (ms): [15918, 15750, 15383, 15708, 15540, 15530, 15424,
15709, 15688, 10542]
2019-06-13 14:00:31,145 INFO [main] hbase.PerformanceEvaluation: [ScanTest
duration ] Min: 10542ms Max: 15918ms Avg: 15119ms
2019-06-13 14:00:31,145 INFO [main] hbase.PerformanceEvaluation: [ Avg latency
(us)] 15
2019-06-13 14:00:31,145 INFO [main] hbase.PerformanceEvaluation: [ Avg
TPS/QPS] 670818 row per second
{noformat}
About 33% faster.
And for scanRange10
{noformat}
./bin/hbase pe --nomapred --rows=1000000 scanRange10 10 &>test_scan10.log <
/dev/null &
{noformat}
master
{noformat}
2019-06-13 14:15:39,578 INFO [main] hbase.PerformanceEvaluation:
[RandomScanWithRange10Test] Summary of timings (ms): [431574, 431034, 426640,
436114, 429464, 431428, 434035, 428176, 430324, 431243]
2019-06-13 14:15:39,910 INFO [main] hbase.PerformanceEvaluation:
[RandomScanWithRange10Test duration ] Min: 426640ms Max: 436114ms Avg:
431003ms
2019-06-13 14:15:39,910 INFO [main] hbase.PerformanceEvaluation: [ Avg latency
(us)] 428
2019-06-13 14:15:39,910 INFO [main] hbase.PerformanceEvaluation: [ Avg
TPS/QPS] 23203 row per second
{noformat}
HBASE-21512
{noformat}
2019-06-13 14:07:42,016 INFO [main] hbase.PerformanceEvaluation:
[RandomScanWithRange10Test] Summary of timings (ms): [364311, 365391, 365243,
367459, 365778, 366567, 364024, 363931, 363785, 366916]
2019-06-13 14:07:42,350 INFO [main] hbase.PerformanceEvaluation:
[RandomScanWithRange10Test duration ] Min: 363785ms Max: 367459ms Avg:
365340ms
2019-06-13 14:07:42,350 INFO [main] hbase.PerformanceEvaluation: [ Avg latency
(us)] 362
2019-06-13 14:07:42,350 INFO [main] hbase.PerformanceEvaluation: [ Avg
TPS/QPS] 27372 row per second
{noformat}
About 18% faster.
I think the reason here is that now the scanner is 'async prefetch' by default.
And I've also tested the old async prefetch scanner, but the result is worse...
{noformat}
./bin/hbase pe --nomapred --rows=1000000 --asyncPrefetch scan 10
&>test_scan_async.log < /dev/null &
{noformat}
{noformat}
2019-06-13 14:27:54,801 INFO [main] hbase.PerformanceEvaluation: [ScanTest]
Summary of timings (ms): [21498, 21307, 20792, 21525, 21006, 21569, 21109,
20920, 21774, 14307]
2019-06-13 14:27:54,858 INFO [main] hbase.PerformanceEvaluation: [ScanTest
duration ] Min: 14307ms Max: 21774ms Avg: 20580ms
2019-06-13 14:27:54,858 INFO [main] hbase.PerformanceEvaluation: [ Avg latency
(us)] 20
2019-06-13 14:27:54,858 INFO [main] hbase.PerformanceEvaluation: [ Avg
TPS/QPS] 492967 row per second
{noformat}
Now, let me focus on the performance for buffered mutator.
> Compare performance number with old sync client implementation
> --------------------------------------------------------------
>
> Key: HBASE-22564
> URL: https://issues.apache.org/jira/browse/HBASE-22564
> Project: HBase
> Issue Type: Sub-task
> Reporter: Duo Zhang
> Assignee: Duo Zhang
> Priority: Major
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
