ignite-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ivan Veselovsky (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (IGNITE-2391) IGFS + Ignite MR works slower than Hadoop (Hortonswork) MR
Date Mon, 25 Jan 2016 10:00:47 GMT

    [ https://issues.apache.org/jira/browse/IGNITE-2391?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15114963#comment-15114963
] 

Ivan Veselovsky edited comment on IGNITE-2391 at 1/25/16 10:00 AM:
-------------------------------------------------------------------

At least one problem in Ignite was found: JIT code cache overflows if the cache size is small,
what causes significant performance degradation. 
To solve this problem we recommend to configure JIT code cache size proportional to the PermGen
size used.
If the problem persists, we'll continue investigation.



was (Author: iveselovskiy):
At least one problem in Ignite was found: JIT code cache overflows if the cache size is small,
what causes significant performance degradation. 
To solve this problem we recommend the customer to configure JIT code cache size proportional
to the PermGen size used.
If the customer encountered the same problem, we'll give him a workaround. If not, we'll continue
the investigation.


> IGFS + Ignite MR works slower than Hadoop (Hortonswork) MR
> ----------------------------------------------------------
>
>                 Key: IGNITE-2391
>                 URL: https://issues.apache.org/jira/browse/IGNITE-2391
>             Project: Ignite
>          Issue Type: Bug
>          Components: IGFS
>    Affects Versions: 1.5
>            Reporter: Denis Magda
>            Assignee: Vladimir Ozerov
>            Priority: Blocker
>              Labels: important
>             Fix For: 1.6
>
>         Attachments: default-config.xml
>
>
> A performance test of IGFS MR vs Traditional HDFS (Hortonswork) MR and performance here
was actually worse using IGFS MR.
> 5 ignite nodes on the 5 data nodes were setup, each also with 64GB ram.
> The same sample search via Hunk was done, making sure Hunk initiated the MR stage. The
only change in both runs was only to run IGFS MR or Traditional MR over HDFS
>  
> Traditional MR over HDFS.
> This search has completed in 54.701seconds.
> IGFS MR
> This search has completed in 75.959seconds.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message