> - The amount of storage needed, should roughly equal the total amount > off traffic generated by this community for one week. > This also puts a requirement on the needed phys mem. the box > should have (see FAQ). It's reading the FAQ that I supposed that in the worst case I need 400 gigabytes of cache storage, and (about) 10 gigabyte of phys. mem. On the physical memory, I'm not so sure, because in the FAQ I read about 10 MB for every GB of disk cache storage, plus what is needed for cache_mem, plus tha RAM used by the OS to cache disk IO. From other sources I read 32MB per GB of disk storage. I don't want to underestimate the need for physical RAM, so I'm taking the "worst" case. I just don't know, and consequently I wonder, if Linux+Squid scales well to this amount of RAM and disk. > - On average usage 12.000 users could lead to a 300reqs/sec range , on > average, which is rather high-end. > I would advise a low-end server with highest cpu-Ghz available. > In that case I would probably use 2 , with load balancing. Do you think that LVS would be a good choice for load balancing? And the servers (which can also be more than 2, if it is advisable) should form a cache array? This should give 2 benefits: if a client requests an object that is in the other server's cache, it is retrieved from there and not from the Internet; and the amount of cache storage should be reduced by roughly a factor of 2. Thank you for your answer.
