hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Chen Qiang (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-9680) Doing a lsr against WebImageViewer is slow
Date Fri, 22 Jan 2016 00:04:40 GMT

    [ https://issues.apache.org/jira/browse/HDFS-9680?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15111642#comment-15111642

Chen Qiang commented on HDFS-9680:

To correct the description:

The fsimage is about 3GB in size with 27051583 files and directories, 26374397 blocks.  It
took ~12 hours to do a '/' lsr dump against WebImageViewer. It takes ~15 mins to run full
lsr against live cluster. 

The WebImageViewer was running with 32GB Memory Heap.

> Doing a lsr against WebImageViewer is slow
> ------------------------------------------
>                 Key: HDFS-9680
>                 URL: https://issues.apache.org/jira/browse/HDFS-9680
>             Project: Hadoop HDFS
>          Issue Type: Improvement
>            Reporter: Haohui Mai
> We have experienced a performance issue that doing lsr against the WebImageViewer.
> For a fsimage that has around 140m files, it takes around ~35 minutes to do the lsr across
the live cluster, but ~12 hours to do the same operation against the WebImageViewer.
> I believe that the root cause is that WebImageViewer decodes the protobuf messages on-demand
which creates a lot of GC pressure. It might be better to decode it at the very beginning.

This message was sent by Atlassian JIRA

View raw message