[LinuxFailSafe] Scalability of Failsafe

Padmanabhan Sreenivasan paddy@sgi.com
Wed, 23 Apr 2003 10:30:55 -0700


> Dileep Matheri wrote:
> 
> Hello All,
> 
> By browsing the Failsafe code, we understood that it supports 32 nodes.
> 
> Right now we are trying to evaluate some linux HA products. One of them is Failsafe.
> The requirement is 32 node support at this point. But in future it may need to scale
> upto 300 nodes. Our questions to the linux Failsafe community is:
> 
> 1. Is Failsafe is scalable beyond 32 nodes?

Not without changes. Some algorithms do not scale well.

> 2. Upto how many nodes Failsafe can scale?

It can work upto 32 nodes (there are assumptions in the
code about the max size of cluster being 32). Linux FailSafe 
has been tested upto 8 nodes.

> 3. If Failsafe can not scale more than 32, what could be the reason for that? or what
>    are the problematic areas?

Cluster membership services algorithm has to be changed. This
algorithm will be the first bottleneck.

Paddy
> Thanks in advance,
> Santhosh/Dileep.