[jira] [Commented] (IGNITE-6108) Hangs in streamer when using store

["Igor Seliverstov (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/IGNITE-6108?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16138293#comment-16138293 ] 

Igor Seliverstov commented on IGNITE-6108:
------------------------------------------

There are two possible causes of slowdowns:
1) swap partition, which starts to be used in case of incorrect configuration (ignite is trying to use more memory than available on the server, some of used memory is allocated in swap partition what breaks overall performance)
2) rare or long checlpoints - if the checkpoint buffer became overflowed, ignate cannot perform update operations in parallel with checkpoints, it has to wait for current checkpoint finish to swap checkpoint buffer.

General recommendations are:

1) use 4 kb pages - it will significantly speed up write operations
2) use several checkpoint threads (4)
3) increase checkpoints frequency not to collect dirty pages and prevent checkpoint buffer overflow


> Hangs in streamer when using store
> ----------------------------------
>
>                 Key: IGNITE-6108
>                 URL: https://issues.apache.org/jira/browse/IGNITE-6108
>             Project: Ignite
>          Issue Type: Bug
>    Affects Versions: 2.1
>            Reporter: Aleksey Chetaev
>            Assignee: Igor Seliverstov
>            Priority: Critical
>         Attachments: logs_without memory_policy.zip, log_with_memory_policy.zip
>
>
> I have hangs in IgniteStreamer benchmarks when using store with WAL mode different from NONE. 
> Configuration: 1 client 4 servers.Cache mode:  transaction. Range 400_000_000. 
> Problems: 
> * Time for next 100_000_000 entries every is always greater than for the previous one.
> * We can found hangs for greater that 1 minutes in drivers logs. E.x. 11:47:21 - 11:48:27 in archive named "log_with_memory_policy".
> Time for every 100_000_000 in seconds in log named "logs_without memory_policy". 
> * 100_000_000 = 209
> * 200_000_000 = 323
> * 300_000_000 = 547
> * 400_000_000 = 722
> Yardstick logs in attachments. 



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