How do squid caches compare with content engines from Cisco? I'm seeking information about the ability of Squid/Linux web caches to handle the caching requirements of a multi-site networked enterprise. Our proposed topology is a hierarchical cache system, with high-end cache servers at the core of the network and next to the internet gateway. Lower end devices would serve the remote sites and some of the intermediate campus locations with moderate traffic. For cost reasons we are considering squid caching software on already purchased computers at these sites.
The network is Cisco-centric, so WCCPv2 is available and recommended for redirecting port 80 traffic. Most cache implementations I've read about redirect with iptables on Linux routers or WCCPv1 on Cisco routers. I want the flexibility of using WCCPv2 if possible, allowing the use of redundant WAN routers that both redirect to the cache(s), as well as the ability to cache web traffic other than port 80. Cisco proposed WCCPv2 as an internet draft standard in July 2000. Does this mean that no licensing from Cisco is necessary for squid releases that support this protocol? I don't have enough squid experience to estimate the cost and effort of deploying and maintaining squid devices as compared to some of the dedicated Cisco cache appliances. I'm wary of being trapped by the false economy of near zero hardware/software costs if ongoing labor costs for maintenance is a major issue. Another concern is performance. Given comparable disk and memory, is it reasonable to specify squid cache hardware that can handle the 2000 simultaneous TCP connections advertised for one of the lower performance CISCO content engines? Would any special linux kernel tuning be neccessary? TIA for any advice. Tom -- ==++==++==++==++==++==++==++==++==++==++==++==++==++==++== Thomas Benjamin benjamin at cactus dot org
