VPLEX: How VPLEX handles backend array QoS settings

Article Number: 514826                             Article Version:                               Article Type: How To


VPLEX Series

VPLEX VS6,VPLEX Series,VPLEX for All Flash,VPLEX GeoSynchrony,VPLEX Local,VPLEX Metro,VPLEX VS2,VPLEX GeoSynchrony 6.0 Service Pack 1 Patch 5,VPLEX GeoSynchrony 6.0 Service Pack 1 Patch 4,VPLEX GeoSynchrony 6.0 Patch 1



VNX, VMAX and Unity has a feature called Host I/O Limits (aka Quality of Service, or QoS)

To limit I/O throughput and bandwidth, this provides more predictable performance in system workloads between hosts, applications and storage resources.



A storage administrator may want to control the amount of I/O requests from non-essential workloads so that these workloads do not tie-up valuable storage resources and affect critical workloads.


An I/O limit is defined as a maximum threshold using either IOs per second (IOPS), or Kilobytes or Megabytes per second (KBPS or MBPS)

If both thresholds are set, the system limits traffic according to the threshold that is reached first.

An I/O limit policy can be applied to an individual LUN or to a group of LUNs. Only one I/O limit policy can be applied to an individual LUN or a LUN that is a member of a consistency group.


Once the maximum limit is reached, I/Os are delayed (queued).


When LUNs with QoS are presented to VPlex along with LUNs with no QoS and then they are presented to a host, only the server using LUNs restricted by the host I/O limit would be affected.


All other servers would continue severing the I/Os normally.

Actually they should experience better response times as the servers that were hammering the BE with many I/Os are being constrained.

If the Host starts bursting I/Os, the Unity will not serve the exceeding I/Os resulting in time outs and this would understandably impact the performance of these LUNs.



If you have 2 servers with adjustable I/O throughput and one is set to 1000 IOPs and the other is on 2000 IOPs and both are using LUNs with QoS of 1000 IOPs.


The server complying with the QoS limit would perform much better than the other one even though they both have effective IOPs of 1000 because the one bursting with IOPs consumes CPU cycles in the IO timeout and re-transmit of IOs.


Also if the same host is presented by 2 LUNs one with host QoS and the other with no QoS then each LUN would experience different latency times with no impact on the LUN that doesn't have host limit configured.