FD_SIMPLE is (a) somewhat indeterministic and (b) has never really been used in production.
The current best practices recommendation is to use *both* FD_SOCK *and* FD.
- FD_SOCK will detect immediately whether a server has crashed or not
- FD will act as second line of defense, handling cases like switch or host crashes
FD_SOCK in pre-2.3 doesn't set KEEP_ALIVE by default, so I suggest you look at the 2.3 sources and add this option yourself (setKeepAlive()).
Thanks for the reply, Bela.
As an aside, the wiki on Failure Detection
http://wiki.jboss.org/wiki/Wiki.jsp?page=FDVersusFD_SOCKhas a broken link at the bottom, pointing to the appropriate issue in JIRA. The href has a "." at the end, which makes it no good.
I think it makes more sense for us to upgrade to 2.3 than to alter the source of 2.2.7.
Just to clarify, I mean upgrade our production systems. I'm still going to mess around with 2.2.7 :)
Thanks again for the advice.
Fixed the link, thanks,