You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@bookkeeper.apache.org by "Sijie Guo (JIRA)" <ji...@apache.org> on 2013/11/10 03:27:17 UTC

[jira] [Updated] (BOOKKEEPER-565) Make an option to separate storing entry log files from index files.

     [ https://issues.apache.org/jira/browse/BOOKKEEPER-565?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Sijie Guo updated BOOKKEEPER-565:
---------------------------------

    Attachment: BOOKKEEPER-565.diff

attach a patch allow configuring index dirs.

> Make an option to separate storing entry log files from index files.
> --------------------------------------------------------------------
>
>                 Key: BOOKKEEPER-565
>                 URL: https://issues.apache.org/jira/browse/BOOKKEEPER-565
>             Project: Bookkeeper
>          Issue Type: Sub-task
>          Components: bookkeeper-server
>            Reporter: Sijie Guo
>            Assignee: Sijie Guo
>             Fix For: 4.3.0
>
>         Attachments: BOOKKEEPER-565.diff
>
>
> Currently, the add operation will go throughput ledger storage first and journal later. the latency of an add operation would be affected by two parts, one is ledger disk and the other one is journal disk. And flushing ledger index files might cause random i/o, which would affect adding entries to ledger storage in a higher workload.
> there are several ideas to resolve this problem, one is aggregate the per-file index into several files to make flushing more sequential.
> this jira is a simple solution to separate the disks to avoid disk contention. it is a straightforward way w/ less risk, but not a perfect way which might waste an extra disk and bandwidth. 



--
This message was sent by Atlassian JIRA
(v6.1#6144)