For the specialized hardware balancing, I'm currently a fan of the F5 BigIP's.. just started working with them and have found them much better to work with than Cisco's CSS/CSM offerings.
Software wise, I've run clusters that have been fronted by a very heavily modified LVS and various scripts on the nodes to provide constant updates so it can calculate very precise weightings (based on things like disk, memory and CPU usage as well as more specifics to the application like mails in the queue) and has proved to be very successful.
With LVS we could scale to around 60 nodes in a cluster before it started to struggle with updates and getting a little confused.
|