[
https://issues.apache.org/jira/browse/SOLR-6557?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14146201#comment-14146201
]
Kenji Kikuchi commented on SOLR-6557:
-------------------------------------
Thank you for pointing to the SOLR-6485.
I read the patch. The patch helps my operations.
I think if maxWriteMBPerSec in the SOLR-6485 can be requested from slave
servers it is more helpful. This is because when I add a slave server in the
rack where a master server exists I can use full server to rack switch
bandwidth. But when I add a slave server in the rack where a master server does
not exist, I need to add a bandwidth cap.
> bandwidth cap for large file replication
> ----------------------------------------
>
> Key: SOLR-6557
> URL: https://issues.apache.org/jira/browse/SOLR-6557
> Project: Solr
> Issue Type: Improvement
> Components: replication (java)
> Affects Versions: 5.0, Trunk
> Reporter: Kenji Kikuchi
> Priority: Minor
> Fix For: 5.0, Trunk
>
> Attachments: SOLR-replication_bandwidth.patch
>
>
> Sometimes I need to set up a slave server in the rack where a master
> server does not exist. In this case, our rack to rack bandwidth is often
> saturated during large file transfer, such as initial replication, large
> index file merge and optimization. This impairs our other services. So I
> think a bandwidth cap for large file replication is helpful for large web
> service providers and adds flexibility to our Solr slave server setups.
> Currently I am limiting replication bandwidth by using a tc command on
> the master servers. But to use a tc command, I need to login to an
> on-service master server and add tc related settings to add a new slave
> server because tc command only shapes outbound traffics. So the feature
> of setting up a desired replication bandwidth cap with just one line in
> a new slave configuration file reduces our Solr operations and secures
> the on-service master servers by avoiding the need to login.
> Parsing bandwidth setting in slave solrconfig.xml in ‘bits per
> second' is preferable for me. This is because most of our site operators
> use ‘bits per second' not ‘bytes per second’ in our network monitoring
> metrics.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
