[jboss-jira] [JBoss JIRA] Closed: (JGRP-1044) UNICAST: use actual retransmission values for timeouts

[Bela Ban (JIRA)]

     [ https://jira.jboss.org/jira/browse/JGRP-1044?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Bela Ban closed JGRP-1044.
--------------------------

    Resolution: Duplicate Issue


JGRP-1053

> UNICAST: use actual retransmission values for timeouts
> ------------------------------------------------------
>
>                 Key: JGRP-1044
>                 URL: https://jira.jboss.org/jira/browse/JGRP-1044
>             Project: JGroups
>          Issue Type: Feature Request
>            Reporter: Bela Ban
>            Assignee: Bela Ban
>             Fix For: 2.9
>
>
> In NAKACK, we already have use_stats_for_retransmission. Implement the same in UNICAST:
> Every AckSenderWindow maintains AVG_XMIT_TIME (a double) which is a rolling average of the retransmission times (diff between sending a message and receiving its ACK). Once we have AVG_XMIT_TIME, we schedule retransmission tasks to be AVG_XMIT_TIME + CONST. CONST should be a few millisconds, but could also be 0.
> The rolling average should start with an initial value (say 30ms), and then add the diff between sending of a message and reception of its ACK, for each ACK received, divided by the number of values. More recent values should be weighted higher, so we have a decay of values over time. We could make this simple by having a bounded number of values, e.g. the last 1000 values.
> So, assuming a CONST of 10, if we have a rolling average of 350ms to member B, then we would schedule retransmission of our messages to B to 360ms. If the avg drops down to 30, we'd retransmit in 40. If it increases to 2300, we'd go to 2310 etc

-- 
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