Isilon InsightIQ User Guide

You can also break out this information by node or disk, which yields more useful information than the aggregate values. The graph represents the amount of time that a disk operation remains in the input-output scheduler, not the actual latency of a disk operation (which is part of the total latency in every input-output request). Long times in the scheduler certainly indicate a busy system, and this may indicate the need to look more closely at cluster statistics to identify potential performance bottlenecks. In general, total latencies in the input-output system should be under 10 milliseconds.