[
https://issues.apache.org/jira/browse/TS-412?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12906824#action_12906824
]
Zhao Yongming commented on TS-412:
----------------------------------
base on the hash functions in iocore/cluster/ClusterHash.cc. I setup a testing
program to get the side effect when one of the machines down, which will
trigger up build_hash_table, and make some hash_table bucket changed.
I use 192.168.0.{1..10} for 10 servers, for example. we count for average
percent.
machines hash_changed
10 0.13%
32 0.62%
64 0.71%
128 0.83%
254 1.23%
so far, as we build the remap using machine IP as a seed for the rand
functions, the IP space will make some small effect. after all, the change is
some is nearly as 1%, described in P_ClusterCache.h: "less than 1% disparity at
255 machines".
the remap design is a great, and we are still seeking some more consistent
solutions if possible, because of xxGb traffic running, we really care of it.
> Cluster hash need to be consist
> -------------------------------
>
> Key: TS-412
> URL: https://issues.apache.org/jira/browse/TS-412
> Project: Traffic Server
> Issue Type: Improvement
> Components: Core
> Environment: all
> Reporter: Zhao Yongming
> Priority: Minor
>
> Currently the cluster hash is not consist, so when some server broken/added,
> all content hash will changed, that is unacceptable when you get BIG cache
> size.
> this is a place hold for consist hash discuss, please comment if you have
> anything concern here.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
