[JBoss JIRA] (ISPN-6027) Use bundler sender-sends-with-timer by default
by Bela Ban (JIRA)
[ https://issues.jboss.org/browse/ISPN-6027?page=com.atlassian.jira.plugin.... ]
Bela Ban edited comment on ISPN-6027 at 1/21/16 7:53 AM:
---------------------------------------------------------
IspnPerfTest with Infinispan 8.2.0-SNAPSHOT
||JGroups || 3.6.5.Final || 3.6.7.Final || 3.6.8-SNAPSHOT ||
|sender-sends-with-timer | 7'500 | 7'700 | 7'800 |
|sender-sends | 10'500 | 10'500 | 10'500 |
|transfer-queue | 11'300 | 11'300 | 11'600 |
|transfer-queue-simplified| - | - | 11'300 |
Bundler type sender-sends-with-timer turns bundlig off, whereas transfer-queue and sender-sends do use bunding.
This table clearly shows that bundling has better performance!
was (Author: belaban):
IspnPerfTest with Infinispan 8.2.0-SNAPSHOT
||JGroups || 3.6.5.Final || 3.6.7.Final || 3.6.8-SNAPSHOT ||
|sender-sends-with-timer | 7'500 | 7'700 | 7'800 |
|sender-sends | 10'500 | 10'500 | 10'500 |
|transfer-queue | 11'300 | 11'300 | 11'600 |
Bundler type sender-sends-with-timer turns bundlig off, whereas transfer-queue and sender-sends do use bunding.
This table clearly shows that bundling has better performance!
> Use bundler sender-sends-with-timer by default
> ----------------------------------------------
>
> Key: ISPN-6027
> URL: https://issues.jboss.org/browse/ISPN-6027
> Project: Infinispan
> Issue Type: Task
> Components: Configuration, Core
> Affects Versions: 8.1.0.Final
> Reporter: Dan Berindei
> Assignee: Dan Berindei
> Fix For: 8.2.0.Beta1
>
> Attachments: ispn-udp.xml, ispn-udp.xml
>
>
> Performance tests show {{sender-sends-with-timer}} is still a bit faster than {{transfer-queue}}, so we should make it the default.
--
This message was sent by Atlassian JIRA
(v6.4.11#64026)
8 years, 11 months
[JBoss JIRA] (ISPN-6027) Use bundler sender-sends-with-timer by default
by Bela Ban (JIRA)
[ https://issues.jboss.org/browse/ISPN-6027?page=com.atlassian.jira.plugin.... ]
Bela Ban commented on ISPN-6027:
--------------------------------
IspnPerfTest with Infinispan 8.2.0-SNAPSHOT
||JGroups || 3.6.5.Final || 3.6.7.Final || 3.6.8-SNAPSHOT ||
|sender-sends-with-timer | 7'500 | 7'700 | 7'800 |
|sender-sends | 10'500 | 10'500 | 10'500 |
|transfer-queue | 11'300 | 11'300 | 11'600 |
Bundler type sender-sends-with-timer turns bundlig off, whereas transfer-queue and sender-sends do use bunding.
This table clearly shows that bundling has better performance!
> Use bundler sender-sends-with-timer by default
> ----------------------------------------------
>
> Key: ISPN-6027
> URL: https://issues.jboss.org/browse/ISPN-6027
> Project: Infinispan
> Issue Type: Task
> Components: Configuration, Core
> Affects Versions: 8.1.0.Final
> Reporter: Dan Berindei
> Assignee: Dan Berindei
> Fix For: 8.2.0.Beta1
>
> Attachments: ispn-udp.xml
>
>
> Performance tests show {{sender-sends-with-timer}} is still a bit faster than {{transfer-queue}}, so we should make it the default.
--
This message was sent by Atlassian JIRA
(v6.4.11#64026)
8 years, 11 months
[JBoss JIRA] (ISPN-6087) Hadoop MapReduce task returns wrong results after one of the servers from the cluster is stopped
by Anna Manukyan (JIRA)
[ https://issues.jboss.org/browse/ISPN-6087?page=com.atlassian.jira.plugin.... ]
Anna Manukyan reassigned ISPN-6087:
-----------------------------------
Assignee: Gustavo Fernandes
> Hadoop MapReduce task returns wrong results after one of the servers from the cluster is stopped
> -------------------------------------------------------------------------------------------------
>
> Key: ISPN-6087
> URL: https://issues.jboss.org/browse/ISPN-6087
> Project: Infinispan
> Issue Type: Bug
> Components: Hadoop Integration
> Reporter: Anna Manukyan
> Assignee: Gustavo Fernandes
> Attachments: FailoverTest.log
>
>
> The test is performed on 4 node cluster. The data read is performed from Infinispan cache, the write is performed to HDFS.
> There are 200000 entries in the cache. As soon as the Hadoop Job is submitted for performing map/reduce task, one of the servers from the cluster is stopped.
> The MapReduce task performs calculation of word count.
> As a result, the Hadoop Map/Reduce result map contains wrong results, that is the number of the certain words found is greater than it is expected.
> The failure doesn't always happen. It depends on which stage was the mapreduce task when the server stopped.
> You can find the test execution log attached.
--
This message was sent by Atlassian JIRA
(v6.4.11#64026)
8 years, 11 months
[JBoss JIRA] (ISPN-6027) Use bundler sender-sends-with-timer by default
by Bela Ban (JIRA)
[ https://issues.jboss.org/browse/ISPN-6027?page=com.atlassian.jira.plugin.... ]
Bela Ban commented on ISPN-6027:
--------------------------------
IspnPerfTest run with Infinispan's UDP config yielded ca. 7500 reqs/sec/node. Changing bundler_type from "sender-sends-with-timer" (effectively turning bundling off) to "transfer-queue" increased perf from 7500 -> 11'000. This was with JGroups 3.6.8-SNAPSHOT and Infinispan 'Chakra' 8.2.0-SNAPSHOT.
If you can come up with a test that shows sender-sends-wth-timer is faster, please post the details here, so I can try to reproduce this on my end.
> Use bundler sender-sends-with-timer by default
> ----------------------------------------------
>
> Key: ISPN-6027
> URL: https://issues.jboss.org/browse/ISPN-6027
> Project: Infinispan
> Issue Type: Task
> Components: Configuration, Core
> Affects Versions: 8.1.0.Final
> Reporter: Dan Berindei
> Assignee: Dan Berindei
> Fix For: 8.2.0.Beta1
>
>
> Performance tests show {{sender-sends-with-timer}} is still a bit faster than {{transfer-queue}}, so we should make it the default.
--
This message was sent by Atlassian JIRA
(v6.4.11#64026)
8 years, 11 months