[
https://issues.apache.org/jira/browse/HBASE-17917?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15994770#comment-15994770
]
Duo Zhang commented on HBASE-17917:
-----------------------------------
The test command is
{noformat}
./bin/hbase pe --rows=100000 --cacheBlocks=false --caching=10
--scanReadType=pread/stream/default --nomapred scanRange10 10
{noformat}
|| ReadType || %us || RS %CPU || DN % CPU ||
| pread | 20 | 450 | 250 |
| stream | 14 | 340 | 130 |
| default | 20 | 450 | 250 |
Streaming read has a lower cpu usage but also a much longer time to finish the
test (1190s vs. 150s). And in fact, after the patch here, we will create a new
DFSInputStream when openning a new scanner for streaming read, so almost every
request to RS will lead to a request to NN. It is not scalable so the bad
performance is expected...
> Use pread by default for all user scan
> --------------------------------------
>
> Key: HBASE-17917
> URL: https://issues.apache.org/jira/browse/HBASE-17917
> Project: HBase
> Issue Type: Sub-task
> Components: scan
> Affects Versions: 2.0.0
> Reporter: Duo Zhang
> Assignee: Duo Zhang
> Fix For: 2.0.0
>
> Attachments: HBASE-17917.patch, HBASE-17917-v1.patch,
> HBASE-17917-v2.patch, HBASE-17917-v2.patch, HBASE-17917-v3.patch,
> HBASE-17917-v4.patch, HBASE-17917-v5.patch
>
>
> As said in the parent issue. We need some benchmark here first.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
