I do not have enough data to say with certainty it is linear. In fact, n+1000 is my rule of thumb which fortunatley has been spot on so far. I never approached it from I/O per agent perspective. Most of my effort is Console centric since my contention is the console introducees more load than any one agent. Also, the n+1000 is for the maximum, the I/O needed under maximum load not average. One solution you may consider, if the env' is on-line, is measure I/O with no agents then with 1 agent, then 2 so on and so forth. However, all agents are not equal. An exchange server will contribute more than a plain ol' file server since the Exchage MP has far more rules and monitors. I am actually in an environment where I am doing this for WIndows 2008 domain controllers. I will post as soon as I have numbers. Sorry I could not give you a more precise answer.