[CTDB and GFS] test: scenario and some results

[Andrew Morgan]

On Fri, 13 Jul 2007, Ignacio Coupeau wrote:

> Volker Lendecke wrote:
>> On Fri, Jul 13, 2007 at 01:36:58PM +0200, Ignacio Coupeau wrote:
>>> I tested the SAMBA cluster CTDB with GFS6.1. The main purpose of this
>>> test was the LB ability of the CTDB.
>>> 
>>> The numeric results are very poor with concurrence (as usual with GFS).
>>> The load balancing issue: well for CTDB, bad for my hardware.
>> 
>> Thanks for testing this!
>> 
>> How do you want to proceed from here?
>
> As our main target is the high availability and load balance, I going to test 
> now with the Nortel 2216 L7 switch instead ServerIron.

We use several different Foundry ServerIrons here, including the XL (not 
for Samba though).

>
> I need check that:
>
> 1. the client traffic are balanced (sticky) to the same node in the whole 
> session (or we have a very bad business with the performance);
>
> 2. if a samba node stops (not only if the hardware stops) the traffic is 
> redirected to another samba node;

Can you post your ServerIron config (sanitized)?  It is certainly possible 
to have the ServerIron do a layer 4 health check on the ports used by 
Samba, which should be sufficient for failover unless Samba is still 
accepting connections but not responding at layer 7.

We use ServerIrons for load balancing a variety of applications, including 
ports that are not well-known by the ServerIron, so hopefully I can help 
you there.

 	Andy