[jboss-jira] [JBoss JIRA] Issue Comment Edited: (JGRP-1315) Discovery: don't send view back with discovery response
[
https://issues.jboss.org/browse/JGRP-1315?page=com.atlassian.jira.plugin....
]
Bela Ban edited comment on JGRP-1315 at 8/15/11 2:06 AM:
---------------------------------------------------------
For discovery requests triggered by GMS (JOIN), every member returns:
- Its own address (UUID)
- Its logical name
- Its physical address
- The coordinator's address (ViewId - creator)
For discovery requests triggered by MERGE2, every member (whose ViewId differs) returns:
- View
For a merge-triggered discovery request, the merge cooordinator sends its *ViewId* with
the request. Only members who have *differing* ViewIds return their *View*.
--> Say we have 3 partitions
--> If we have 100 members, the existing code sends 100 discovery responses, whereas
the new code would send 2 responses !
Further optimization:
- Add a staggering timeout:
--> Every member picks a random value in the range [0..timeout] and waits that amount
of timeout before it sends the response. This will prevent traffic bursts. Timeout=0 means
don't stagger the response.
was (Author: belaban):
For discovery requests triggered by GMS (JOIN), every member returns:
- Its own address (UUID)
- Its logical name
- Its physical address
- The coordinator's address (UUID)
For discovery requests triggered by MERGE2, every member (whose ViewId differs) returns:
- View
However, for a merge-triggered discovery request, the merge cooordinator sends its
*ViewId* with the request. Only members who have *differing* ViewIds return their View.
--> Say we have 3 partitions
--> If we have 100 members, the existing code sends 100 discovery responses, whereas
the new code would send 2 responses !
--> We do not need to send the entire View, which saves space on large clusters
Further optimization:
- Add a staggering timeout:
--> Every member picks a random value in the range [0..timeout] and waits that amount
of timeout before it sends the response. This will prevent traffic bursts. Timeout=0 means
don't stagger the response.
Discovery: don't send view back with discovery response
-------------------------------------------------------
Key: JGRP-1315
URL: https://issues.jboss.org/browse/JGRP-1315
Project: JGroups
Issue Type: Feature Request
Reporter: Bela Ban
Assignee: Bela Ban
Fix For: 2.12.2, 3.0
We send back a View with each discovery response. This is wasteful if we have large
clusters.
SOLUTION: change this to only return the coordinator's address for initial
discovery.
Note: look into only returning the ViewId for merge triggered discovery
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira