cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jeff Liu (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-10714) tcp retransmission issue seen in cassandra cluster
Date Tue, 17 Nov 2015 18:41:11 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-10714?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15009217#comment-15009217
] 

Jeff Liu commented on CASSANDRA-10714:
--------------------------------------

[~mshuler], I checked the NIC status and it shows errors:0 dropped:0 overruns:0 frame:0 on
both rx and tx. I even increased the tx queue size to 5000 from default 1000 without any improvement.
Specifically for linux networking settings, I also increased both net.core.rmem, net.core.wmem,,
net.ipv4.tcp_rmem, net.ipv4.tcp_wmem to 32M. increasd net.core.netdev_max_backlog to 30000
with no luck to eliminate the retransmission. 

> tcp retransmission issue seen in cassandra cluster
> --------------------------------------------------
>
>                 Key: CASSANDRA-10714
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-10714
>             Project: Cassandra
>          Issue Type: Improvement
>          Components: Streaming and Messaging
>            Reporter: Jeff Liu
>
> I have been seen tcp package retransmission issue in various stacks in our environment.
( AWS with no VPC). I'm currently using hsha rpc server type on cassandra 2.1.6 version. The
information captured by wireshark shows that the retransmission happened both between client-to-server
and server-to-server. Even within a cluster that doesn't have any client traffic, sporadic
retransmission still happens.
> It's pretty easy to reproduce this issue by watching "netstat -s | grep retrans". 



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message