Redhat 2.1 kernel tcp timeout settings for RAC

Hi, got a 4 node 9.2.0.5 cluster on RH 2.1, we have to take one machine out of action for some hrware issues. Having some trouble with the network timeout settings, with the oracle services / listener down but the machine up any connections get re-routed instantanoiusly to the other node which is what we want.

When the 1 machine is powered off which it needs to be then connections take 20 seconds to be mae - this ia a bit too long, want to get it down to 2 - 3 seconds at the very maximum.

The 20 seconds matches what is in metalink note 249213.1 - but we need it quicker

I am guessing this is TCP timeouts in the kernel, the current settings for that are (these are the CLIENT settings)

net.ipv4.tcp_keepalive_time = 30
net.ipv4.tcp_keepalive_intvl = 60
net.ipv4.tcp_retries2 = 3
net.ipv4.tcp_syn_retries = 2

Anyone got any suggestion on what i can put these to? Or am I missing something else

Thanks

Dave

--
http://www.freelists.org/webpage/oracle-l