[infinispan-issues] [JBoss JIRA] (ISPN-5083) Hot Rod decoder should use async Cache operations
Radim Vansa (JIRA)
issues at jboss.org
Thu Dec 18 09:12:29 EST 2014
[ https://issues.jboss.org/browse/ISPN-5083?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13028669#comment-13028669 ]
Radim Vansa commented on ISPN-5083:
-----------------------------------
Maybe I realized what you might mean by this. Do you expect that a thread will read the request, issue asynchronous operation and the thread handling this async operation will later write the response to the socket? That would allow server-side multiplexing of several HotRod RPC calls over one connection, which is definitely a good thing! Though, in order to fully utilize such power, you'd need the client to multiplex the calls over this single socket, too. This is what SpyMemcached client does, IIRC, and our server implementation prohibits efficient use of socket connections due to the fact that the connection is owned by thread processing the HotRod call.
> Hot Rod decoder should use async Cache operations
> -------------------------------------------------
>
> Key: ISPN-5083
> URL: https://issues.jboss.org/browse/ISPN-5083
> Project: Infinispan
> Issue Type: Enhancement
> Components: Remote Protocols
> Reporter: Galder Zamarreño
> Assignee: Gustavo Fernandes
> Fix For: 7.1.0.Final
>
>
> Hot Rod decoder is currently tying up Netty threads as a result of calling up to Infinispan sync operations. Instead, Hot Rod decoder should call up async operations, convert the Notifying Futures to Scala Futures, and write up the reply when it's received. This should increase performance specially under heavy load.
--
This message was sent by Atlassian JIRA
(v6.3.11#6341)
More information about the infinispan-issues
mailing list