[ https://jira.jboss.org/jira/browse/JGRP-100?page=com.atlassian.jira.plugi... ] Bela Ban commented on JGRP-100: ------------------------------- Further points to look at: GMS: - Is the single coordinator a bottle neck ? When a coord hangs, it won't be able to admit new nodes into the cluster, until it is suspected and excluded - Use view_bundling Failure detection: - Replace FD with FD_ALL TP, timers: - Increase thread pools NAKACK: - Set use_mcast_xmit="true" if UDP is used. UDP is recommended anyway for large clusters - Use stats for xmit_timeouts ? PBCAST: - Revisit ? SMACK: - Use instead of NAKACK ? Doesn't require GMS - Improve: don't ack every single message ! FC: - Increase max_credits FRAG2: - Use instead of FRAG - Get latest version (fix for https://jira.jboss.org/jira/browse/JGRP-800) Streaming API (?): - For large values ? Discovery: - Use only servers, not clients (https://jira.jboss.org/jira/browse/JGRP-735) - Workaround: set num_responses to a large value, so we do get some server responses -- This message is automatically generated by JIRA. - If you think it was sent incorrectly contact one of the administrators: https://jira.jboss.org/jira/secure/Administrators.jspa - For more information on JIRA, see: http://www.atlassian.com/software/jira