[
https://issues.jboss.org/browse/JGRP-1722?page=com.atlassian.jira.plugin....
]
Bela Ban commented on JGRP-1722:
--------------------------------
As every message that needs to be encrypted (when sent or received) has to acquire the
*single* lock in {{ENCRYPT}}, even unrelated messages (e.g. from different senders or
tagged as {{OOB|DONT_BUNDLE}}) are delivered or sent sequentially.
SOLUTION: use a pool of ciphers. Each cipher has a related lock. A sender or receiver
thread grabs a random cipher and locks its associated lock. This way, lock requests are
striped and the single lock bottleneck is removed. The pool size (and thus the lock
striping) needs to be configurable.
Improve performance of ENCRYPT protocol
---------------------------------------
Key: JGRP-1722
URL:
https://issues.jboss.org/browse/JGRP-1722
Project: JGroups
Issue Type: Enhancement
Affects Versions: 3.4
Reporter: Martin Gencur
Assignee: Bela Ban
Fix For: 3.5
A stress tests with the following setup showed that performance (reads and writes/sec) is
halved when ENCRYPT protocol is enabled:
Infinispan had distributed sync cache with 2 owners on 4 nodes, no transactions. The
stress test used 10 threads on each node accessing 1024 byte entries, no conflicts on
keys, 20 % writes, 80 % reads.
It would be great if we could improve the performance of ENCRYPT protocol.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see:
http://www.atlassian.com/software/jira