Metrics to calculate 'best' values for cache_mem and cache_dir?

[Peter Gaughran <Peter.Gaughran@xxxxxxx>]

With the recent (enlightening) posts from Christian, Marcus et al, I was 
wondering what the metrics actually are for calculating cache_mem and 
cache_dir? I've read (for Linux at least) aufs is superior to diskd, so 
I've made the appropriate decision, but any time any reference is made 
to size of caches or directory depth, figures are mentioned that seem to 
be obeying formulae, but I've yet to find the actual formulae? Are they 
any other critical parameters I should be wary of, resource wise?

I've had a look over the Squid wiki, and various other sources, but 
can't seem to find a definitive answer? Apologies if I'm being daft and 
missing something that's well known :)

Our current setup is a 32 gig, single CPU (E5620) with 4 15k 300 gig 
disks available for disk cache (o/s is elsewhere).