Tuning Round Robin IOPS Limit

Quick Answer

Increasing the Round Robin IOPS limit to 10000 is correct because the default value of 1000 was causing the path selection policy to switch between paths far more often than the workload needed. Round Robin distributes I/O across all active paths, but the IOPS limit setting controls how many I/O operations are sent down one path before the host switches to the next; a low limit like 1000 means the host is constantly re-selecting paths, and that switching overhead becomes especially costly for high-I/O VMs generating heavy, continuous traffic. Since the storage administrator confirmed the array is genuinely active-active under ALUA, all paths are equally capable of serving I/O efficiently, so there is no reason to keep switching so aggressively; raising the limit lets each path carry substantially more traffic before a switch occurs, reducing churn and letting the array's active-active capability actually be used efficiently. This differs from a scenario involving an active-passive or asymmetric array, where frequent path switching or the wrong PSP entirely would be the real problem. When a question describes degraded throughput on a Round Robin-configured, active-active datastore with paths reporting healthy but inconsistent latency, and the IOPS limit is still at its low default, suspect that the switching frequency itself, not the paths, the SATP, or the array, is the bottleneck.