ignite-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ivan Rakov (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (IGNITE-5961) Align pages in LFS partition files to pageSize
Date Mon, 07 Aug 2017 14:20:00 GMT

     [ https://issues.apache.org/jira/browse/IGNITE-5961?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Ivan Rakov updated IGNITE-5961:
-------------------------------
    Description: 
We store 17 bytes of header at the start of every partition file:
{noformat}
    /** Allocated field offset. */
    static final int HEADER_SIZE = 8/*SIGNATURE*/ + 4/*VERSION*/ + 1/*type*/ + 4/*page size*/;
{noformat}
Even if pageSize is equal to OS page cache size and equal to SSD disk page size (which is
best scenario), we generate two dirty pages instead of one. This is suboptimal and can be
a bottleneck of checkpoint write speed. 

  was:
We store 17 bytes of header at the start of every partition file:
{noformat}
    /** Allocated field offset. */
    static final int HEADER_SIZE = 8/*SIGNATURE*/ + 4/*VERSION*/ + 1/*type*/ + 4/*page size*/;
{noformat}
Even if pageSize is equal to OS page cache size and equal to SSD disk page size (which is
best scenario), when generate two dirty pages insteadf of one. This is suboptimal and can
be a bottleneck of checkpoint write speed. 


> Align pages in LFS partition files to pageSize
> ----------------------------------------------
>
>                 Key: IGNITE-5961
>                 URL: https://issues.apache.org/jira/browse/IGNITE-5961
>             Project: Ignite
>          Issue Type: Improvement
>          Components: persistence
>    Affects Versions: 2.1
>            Reporter: Ivan Rakov
>            Assignee: Ivan Rakov
>            Priority: Critical
>             Fix For: 2.2
>
>
> We store 17 bytes of header at the start of every partition file:
> {noformat}
>     /** Allocated field offset. */
>     static final int HEADER_SIZE = 8/*SIGNATURE*/ + 4/*VERSION*/ + 1/*type*/ + 4/*page
size*/;
> {noformat}
> Even if pageSize is equal to OS page cache size and equal to SSD disk page size (which
is best scenario), we generate two dirty pages instead of one. This is suboptimal and can
be a bottleneck of checkpoint write speed. 



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Mime
View raw message