Release Notes

17 Analyzing Dell PS Series Storage with SAN Headquarters | TR1050
• Throughput: Typically measured in megabytes per second (MB/s), this is the amount of bandwidth needed for
the application. This may be derived by multiplying the I/O size by the number of IOPS. Understanding
saturation points at the port level is important for high bandwidth needs.
• Queue depth: Indicates the number of requests that are lined up waiting on processing.
• Latency: The amount of time a request takes to arrive at the application. Typically, latency is the most telling
symptom of an issue.
2.4 Performance analysis overview
For the purpose of this document, the focus will be on the scope that SAN Headquarters provides. This will include the
storage virtualization layers, the IOPS to the volumes, pools, members and groups. In addition, the backend IOPS to the
actual disks within each member will be considered.
2.4.1 Hardware considerations
Hardware issues should be resolved appropriately and all connectivity should follow best practices. Documentation
that addresses best practices for switches, RAID selection, and advanced PS Series features is available on PS Series
technical documents and videos.
The following methodology may help you identify most PS Series storage performance issues:
• Review the overall environment.
• Compare the actual I/O load to the expected theoretical maximum.
• Determine the actual saturation level of the disks by comparing the observed I/O with the rule-of-thumb
expectation.
• Verify that there is ample capacity available for load balancing, snapshots, and reserve space.
The overall goal is to find the bottleneck that is contributing to unacceptable responsiveness.
2.4.2 Network considerations
When setting up switches and host NICs, following best practices for PS Series storage typically eliminates the network
as the source of a performance issue. Other issues will be apparent if the best practices are not followed.
TCP retransmits or other network issues may be avoided by following these practices:
• Isolate iSCSI traffic.
• Enable Flow Control.
• Ensure all switches are interconnected.
- Non stacked: enable (LAGs) with rapid spanning tree
- Stacked: use uplinks
• Enable Jumbo Frames (optional).