Hey all - I have seen that the rule of thumb for a forward proxy setup to have at least 3 days for the "LRU reference age" in cache mgr's /storedir as a good measure of the working set. But does anyone have any input on what it should be for a reverse-proxy (accelerator) architecture, with respect to the number of objects ? I'd love to see if anyone has any whitepapers or docs that have any information surrounding tuning cache size of an existing production load. I understand it will be dependent on the amount of churn/usage and might vary from site to site, but I'm curious about what list members think. My current plan is to increase slowly the amount of disk cache on each of my boxes and keep a close eye on memory usage, disk I/O, and cache hit times. (Setup is Linux, 6 disk SCSI, each cache_dir on separate spindles, 2048mb cache_mem.) On a related note, I can confirm that the old "10mb memory per 1G disk cache" rule of thumb is about right. :) -john
