Details
-
Improvement
-
Status: Closed
-
Major
-
Resolution: Fixed
-
3.6.0
Description
I found out the JMX executor wrapper of the Netty protocol transport layer was doing locks (synchronized lists and synchronized blocks).
On a regular SMTP worload (5000 mails of 500 KB) we spent 12.9% of CPU cycles doing locks... (See attached flame graphs)
We can easily implement the very same feature without any lock. Wich result in using 4.7 % only of CPU cycles.
Attachments
Attachments
Issue Links
- links to