Analyzing an I/O Bottleneck

An I/O bottleneck means your system's throughput cannot be increased because I/O requests cannot be handled fast enough. If your users do not have this I/O problem on a lightly loaded system, then it must be caused by some form of I/O contention, that is, there must be too many simultaneous requests from different users for service on a device, control unit, or channel path level. Many of your in-queue users will then be found in 'I/O wait' state, and especially their I/O-bound work will take much longer during peak hours. Please realize, however, that the I/O problem might also be restricted to a specific disk. It could then concern only a certain group of users, and the %IQ value would not necessarily indicate the problem.

If, on the other hand, a user feels that an I/O bound job takes too long even when there is no other heavy I/O activity on the system, the problem must be caused by too many I/O requests being made by this user, and this may not lead to I/O contention but just to some I/O devices running at very high rates. You should then try to improve this specific user's I/O environment (see Analyzing a Single Virtual Machine).