You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@cassandra.apache.org by "Jason Brown (JIRA)" <ji...@apache.org> on 2016/07/29 22:11:20 UTC

[jira] [Commented] (CASSANDRA-12346) Gossip 2.0 - introduce a Peer Sampling Service for partial cluster views

    [ https://issues.apache.org/jira/browse/CASSANDRA-12346?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15400097#comment-15400097 ] 

Jason Brown commented on CASSANDRA-12346:
-----------------------------------------

Here is the branch for an implementation of hyparview:

||hyparview||
|[branch|https://github.com/jasobrown/cassandra/tree/broadcast_hyparview]|
|[dtest|http://cassci.datastax.com/view/Dev/view/jasobrown/job/jasobrown-broadcast_hyparview-dtest/]|
|[testall|http://cassci.datastax.com/view/Dev/view/jasobrown/job/jasobrown-broadcast_hyparview-testall/]|

I've documented it rather extensively, so I hope that aides the reviewer. There are still a couple of (very) minor things left to clean up, including the simulator (implemented as a long test), but that should not hinder any review, i think.


> Gossip 2.0 - introduce a Peer Sampling Service for partial cluster views
> ------------------------------------------------------------------------
>
>                 Key: CASSANDRA-12346
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-12346
>             Project: Cassandra
>          Issue Type: Improvement
>          Components: Core
>            Reporter: Jason Brown
>            Assignee: Jason Brown
>              Labels: gossip
>
> A [Peer Sampling Service|infoscience.epfl.ch/record/83409/files/neg--1184036295all.pdf] is a module that provides a partial view of a cluster to dependent modules. A node's partial view, combined with all other nodes' partial views, combine to create a fully-connected mesh over the cluster. This way, a given node does not need to have direct connections to every other node in the cluster, and can be much more efficient in terms of resource management as well as information dissemination. Peer Sampling Services by their nature must be self-healing and self-balancing to maintain the fully-connected mesh.
> I propose we use an algorithm based on [HyParView (http://asc.di.fct.unl.pt/~jleitao/pdf/dsn07-leitao.pdf], which is a concrete algorithm for a Peer Sampling Service. HyParView has a clearly defined protocol, and is reasonably simple to implement.



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