RAID configurations had always been discussed in a server context until we tried to use them at the client end. Multi-user situations must change the value of it.
Indeed. Activity could be, at worst, as much as each client's separate activity combined. With reads, there's usually very little way around this, if caching can't mitigate most of it. So if each <whatever> to service a single user takes 5 read IOPS once over the cache's size, assuming that IO is a negligible problem until the cache is maxed out, and around 100 concurrent users is enough to saturate the cache, then 200 concurrent users might need 500 IOPs. That is of course unrealistically simplistic, but once RAM isn't enough, servers can quickly become storage-limited due to many users each requiring a certain rate of random IOPS served.
Facebook
Twitter