[
https://issues.apache.org/jira/browse/TAJO-269?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14260199#comment-14260199
]
ASF GitHub Bot commented on TAJO-269:
-------------------------------------
Github user hyunsik commented on the pull request:
https://github.com/apache/tajo/pull/322#issuecomment-68272663
I carried out some simple benchmark in terms of serialized data size and
(de)serialization speed. Protobuf-based serialization outperforms both
serialized size and serialization speed. They also have similar deserialization
speed.
Test query:
```
create table store1 as select p.deptName, sumtest(score) from dept as p,
score group by p.deptName.
```
## Size
* JSON serialized size: 9,597 bytes
* Protobuf serialized size: 2,131 bytes
## Speed
I used 10,000 iteration to measure times of serialization and
deserialization.
*Json*
* Serialization: 5,265 msec
* Deserialization; 10,269 msec
*Protobuf*
* Serialization: 1,779 msec
* Deserialization: 10244 msec
> Protocol buffer De/Serialization for LogicalNode
> ------------------------------------------------
>
> Key: TAJO-269
> URL: https://issues.apache.org/jira/browse/TAJO-269
> Project: Tajo
> Issue Type: Improvement
> Components: query master, worker
> Reporter: Jihoon Son
> Assignee: Hyunsik Choi
> Fix For: 0.10
>
> Attachments: TAJO-269.patch, TAJO-269_2.patch
>
>
> In the current implementation, the logical plan is serialized into a JSON
> object and sent to each worker.
> However, the transmission of JSON object incurs the high overhead due to its
> large size.
> ProtocolBuffer is a good alternative because its overhead is quite small and
> already used in other modules of Tajo.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
