[
https://issues.apache.org/jira/browse/HBASE-18485?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16114496#comment-16114496
]
Guanghao Zhang commented on HBASE-18485:
----------------------------------------
bq. That is a potential deadlock.
Nice catch.
bq. Is signal() more suitable than signalAll()?
Yes, there only one thread to call next and another thread to load cache. But I
thought it still should use signalAll() to avoid misuse? And for most cases, we
should use singalAll() instead of signal(). Thanks.
> Performance issue: ClientAsyncPrefetchScanner is slower than
> ClientSimpleScanner
> --------------------------------------------------------------------------------
>
> Key: HBASE-18485
> URL: https://issues.apache.org/jira/browse/HBASE-18485
> Project: HBase
> Issue Type: Improvement
> Affects Versions: 3.0.0, 2.0.0-alpha-2
> Reporter: Guanghao Zhang
> Assignee: Guanghao Zhang
> Fix For: 3.0.0, 2.0.0-alpha-2
>
> Attachments: HBASE-18485-v1.patch, HBASE-18485-v2.patch,
> HBASE-18485-v3.patch, HBASE-18485-v4.patch, HBASE-18485-v4.patch,
> HBASE-18485-v4.patch, HBASE-18485-v5.patch, HBASE-18485-v5.patch,
> HBASE-18485-v6.patch
>
>
> Copied the test result from HBASE-17994.
> {code}
> ./bin/hbase org.apache.hadoop.hbase.PerformanceEvaluation --rows=100000
> --nomapred scan 1
> ./bin/hbase org.apache.hadoop.hbase.PerformanceEvaluation --rows=100000
> --nomapred --asyncPrefetch=True scan 1
> {code}
> Mean latency.
> || ||Test1|| Test2 || Test3 || Test4|| Test5||
> |scan| 12.21 | 14.32 | 13.25 | 13.07 | 11.83 |
> |scan with prefetch=True | 37.36 | 37.88 | 37.56 | 37.66 | 38.28 |
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
