lingu wrote: > Help me in designing the setup of high availability squid I think your better off using a load balancer instead of a cluster. For one, GFS requires shared storage(typically SAN), and two load balancing is much simpler than clustering. LVS is a free linux-based load balancer, so you'd have two LVS nodes and N+1 squid nodes. I prefer F5 BigIPs as they are more flexible/much easier to manage though they aren't that cheap, though if you have 4000 clients it shouldn't be hard to justify it. Run round-robin load balancing, and if your load balancer supports it consider enabling some sort of persistence such as source IP persistence so that the client IPs get pinned to the same proxy server(so the cache is consistent) throughout their session, if that proxy server fails then the load balancer will fail them over to the other server seamlessly. Don't worry about keeping the caches in sync. With BigIP you can configure this in about 30 seconds, with LVS I'm not sure if it's even possible. nate