[ http://jira.jboss.com/jira/browse/JGRP-267?page=comments#action_12340635 ] Bela Ban commented on JGRP-267: ------------------------------- seems that when UDP.loopback is enabled, this works okay... -- This message is automatically generated by JIRA. - If you think it was sent incorrectly contact one of the administrators: http://jira.jboss.com/jira/secure/Administrators.jspa - For more information on JIRA, see: http://www.atlassian.com/software/jira

[ http://jira.jboss.com/jira/browse/JGRP-267?page=comments#action_12340640 ] Bela Ban commented on JGRP-267: ------------------------------- when we instrument the channel.send() method so see how long a send() takes, we get the results below. Most of the time, send() takes almost no time, but sometimes it takes 15ms ! Need to investigate (e.g. with PERF) to see where that time is spent ! $ jt RoundTrip -props ./tmp.xml -num 10 ------------------------------------------------------- GMS: address is 192.168.5.2:2560 ------------------------------------------------------- sending 10 requests [JChannel] send took 0ms - received 1 [JChannel] send took 0ms - received 2 [JChannel] send took 0ms - received 3 [JChannel] send took 0ms - received 4 [JChannel] send took 16ms - received 5 [JChannel] send took 0ms - received 6 [JChannel] send took 15ms - received 7 [JChannel] send took 0ms - received 8 [JChannel] send took 0ms - received 9 [JChannel] send took 0ms - received 10 Took 3156ms for 10 requests: 3.168567807351077 requests/sec, 315.6 ms/request -- This message is automatically generated by JIRA. - If you think it was sent incorrectly contact one of the administrators: http://jira.jboss.com/jira/secure/Administrators.jspa - For more information on JIRA, see: http://www.atlassian.com/software/jira

[ http://jira.jboss.com/jira/browse/JGRP-267?page=comments#action_12340697 ] Bela Ban commented on JGRP-267: ------------------------------- The problem occurs when sending small packets (size: 10 bytes) with DatagramSocket.send(): sometimes it takes 16ms for send() to return. The send buffer should *not* be the problem as it is big enough. When sending larger packets (e.g. 1000 bytes), the problem almost never occurs ! This needs to be further investigated once (if!) we have an APR-based transport, will move to 2.5 and add a dep to APR transport -- This message is automatically generated by JIRA. - If you think it was sent incorrectly contact one of the administrators: http://jira.jboss.com/jira/secure/Administrators.jspa - For more information on JIRA, see: http://www.atlassian.com/software/jira

[ http://jira.jboss.com/jira/browse/JGRP-267?page=comments#action_12340704 ] Bela Ban commented on JGRP-267: ------------------------------- I removed *all* protocols, left only UDP: we have the *same* behavior, so it is not dependent on the other protocols. Meanwhile, I also created RoundTripTcp and RoundTripUdp, which are very fast (see below). Now, I need to find out what is the diff between (1) RoundTrip with just UDP in the config and (2) RoundTripUdp ! bela@dell /cygdrive/c $ jt RoundTripUdp -host 192.168.5.2 -num 1000 sending 1000 requests - received 100 - received 200 - received 300 - received 400 - received 500 - received 600 - received 700 - received 800 - received 900 received all responses (1000) Took 172ms for 1000 requests: 5813.9534883720935 requests/sec, 0.172 ms/request -- This message is automatically generated by JIRA. - If you think it was sent incorrectly contact one of the administrators: http://jira.jboss.com/jira/secure/Administrators.jspa - For more information on JIRA, see: http://www.atlassian.com/software/jira

[ http://jira.jboss.com/jira/browse/JGRP-267?page=all ] Bela Ban updated JGRP-267: -------------------------- Fix Version/s: 2.4 (was: 2.x) -- This message is automatically generated by JIRA. - If you think it was sent incorrectly contact one of the administrators: http://jira.jboss.com/jira/secure/Administrators.jspa - For more information on JIRA, see: http://www.atlassian.com/software/jira