[
https://issues.apache.org/jira/browse/HAMA-526?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14502393#comment-14502393
]
Edward J. Yoon commented on HAMA-526:
-------------------------------------
Another bottleneck is here:
{code}
public void sendMessage(V vertexID, M msg) throws IOException {
peer.send(getHostName(vertexID), new GraphJobMessage(vertexID, msg));
}
{code}
When sendMessage is called, we have to collect values per vertex locally and
create message object at once, instead of create new object everytime.
> Multi-threaded vertex processing
> --------------------------------
>
> Key: HAMA-526
> URL: https://issues.apache.org/jira/browse/HAMA-526
> Project: Hama
> Issue Type: Sub-task
> Components: bsp core, graph
> Reporter: Edward J. Yoon
> Assignee: Edward J. Yoon
> Attachments: patch.txt
>
>
> {code}
> public void bsp(BSPPeer peer) throws IOException, SyncException,
> InterruptedException {
>
> ...
> while (updated && iteration < maxIteration) {
> peer.sync();
> ...
> for (Map.Entry<String, LinkedList<Writable>> e : msgMap.entrySet()) {
> if (e.getValue().size() > 0) {
> vertices.get(e.getKey()).compute(e.getValue().iterator());
> }
> }
> ...
> }
> {code}
> Above code will run Vertex sequentially. Do you think we can run Vertex
> processing concurrently using multi-threads?
> Then what's the advantages or disadvantages of using multi-threading?
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
