[Linux-cluster] Network failure results cluster environment unstable & fragile

[Lon Hohberger]

On Fri, 2006-02-24 at 12:38 +0530, Deval kulshrestha wrote:
> Dear Suvankar
> 
> On my setup, I have to run Mail and Web solution with some additional
> stuff(Mysql, Anti virus, Anti spam etc...).
> I had never tried oracle on this.
> 
> Sorry, 
> but do you have robust cluster environment. For me it's still looking very
> fragile(in network perspective), if network goes-off, or at peak load when
> nodes can not communicate to each other. This makes big chaos.

You need to play with the timers in /proc/cluster

  /proc/cluster/config/cman/hello_timer
  /proc/cluster/config/cman/deadnode_timeout

Increase the value of deadnode_timeout by 20-40 seconds on each of the
cluster nodes, and it should be more resilient to load increases.

-- Lon