[jira] [Updated] (KAFKA-4904) Performance of RocksDb with state record cache

2018-01-26 Thread Guozhang Wang (JIRA)

 [ 
https://issues.apache.org/jira/browse/KAFKA-4904?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Guozhang Wang updated KAFKA-4904:
-
Fix Version/s: (was: 1.1.0)

> Performance of RocksDb with state record cache
> --
>
> Key: KAFKA-4904
> URL: https://issues.apache.org/jira/browse/KAFKA-4904
> Project: Kafka
>  Issue Type: Improvement
>  Components: streams
>Affects Versions: 0.10.2.0
>Reporter: Eno Thereska
>Priority: Major
>
> The performance of RocksDb without any record cache is slightly better than 
> with the default of 10MB of the default cache. This needs investigating. It 
> is likely to be the case that this is not an entirely apples-to-apples 
> comparison since the record cache holds records for a maximum of 
> 'commit.time.ms', which by default is 30 seconds. So the record cache is 
> adding quite a bit of latency, and we know that, however documenting the 
> tradeoff and looking if there is any other bugs needs to be done.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)


[jira] [Updated] (KAFKA-4904) Performance of RocksDb with state record cache

2017-09-22 Thread Guozhang Wang (JIRA)

 [ 
https://issues.apache.org/jira/browse/KAFKA-4904?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Guozhang Wang updated KAFKA-4904:
-
Fix Version/s: (was: 1.0.0)
   1.1.0

> Performance of RocksDb with state record cache
> --
>
> Key: KAFKA-4904
> URL: https://issues.apache.org/jira/browse/KAFKA-4904
> Project: Kafka
>  Issue Type: Improvement
>  Components: streams
>Affects Versions: 0.10.2.0
>Reporter: Eno Thereska
> Fix For: 1.1.0
>
>
> The performance of RocksDb without any record cache is slightly better than 
> with the default of 10MB of the default cache. This needs investigating. It 
> is likely to be the case that this is not an entirely apples-to-apples 
> comparison since the record cache holds records for a maximum of 
> 'commit.time.ms', which by default is 30 seconds. So the record cache is 
> adding quite a bit of latency, and we know that, however documenting the 
> tradeoff and looking if there is any other bugs needs to be done.



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