[Openais] rrp test

[Steven Dake]

On Mon, 2006-08-14 at 17:29 +0200, Fabien THOMAS wrote:
> I've done some test of the ring redundancy code under FreeBSD with  
> the following configuration:
> 
> 3 FreeBSD node with two ring (2 NIC per ring)
> 
> It works fine except for:
> 
> 1) openais-cfgtool completly lock the station until aisexec is  
> exited: removing process priority correct the issue but i dont known  
> why during the socket close the process
> is locked.
> 

Unfortunately all of my systems have multiple processors in my redundant
ring configuration.  It sounds like some kind of spinning in the
executive.  But with my multi-procs, I can't find these sorts of
deadlocks easily.  I'll take a look at this when I have a chance.

> 2) hole (around 5s) before ring failure detection: it seems that  
> during a fault before the ring is declared faultly all communication  
> on the other rings is disabled.
> Is it a known issue ?
> 

Yes, this is how redundant ring works (if you kill a nic interface).
The protocol must synchronize the token rotating around the ring, or it
would be possible for the rings to become out of sync.  The
synchronization is done by waiting for both tokens from both interfaces.
If a token isn't received in a timeout period, its failure counter will
be increased until it reaches a threshold, at which point the ring will
be declared faulty.

It is possible to reduce the timing parameters and reduce the amount of
times a message has failed since receipt to improve this situation.
Keep in mind that the minimum timing parameters of the OS come into play
here, so the best that could probably be done is around 1 sec for
redundant ring failure detection.

In the future, we could likely make improvements on this original design
to improve failure detection time for redundant rings.

Regards
-steve

> Regards,
> Fabien
> 
> 
> 
> 
> _______________________________________________
> Openais mailing list
> Openais at lists.osdl.org
> https://lists.osdl.org/mailman/listinfo/openais