Release Notes

15 Analyzing Dell PS Series Storage with SAN Headquarters | TR1050
Figure 7 When observing latency relationships between components, DDR memory is much faster than spinning
disks
Looking at the previous figure, all resources should be understood from a utilization perspective. If the server has
saturated processors or memory, latency suffers overall. However, if the overall acceptable response time to and from
the storage subsystem is more than the application's tolerance, the disk should be the focus of the investigation.
2.2 Understand the PS Series scope of monitoring
Storage subsystems have improved volume latency by aggregating disks together which effectively scales the total
amount of disks in the set. For PS Series storage, disks are logically grouped together in pools which aggregates all the
disks in that pool to support the application's need for data. For instance, a single 15K disk capable of 180 IOPS will be
able to provide near 3960 IOPS in a 22-disk set (22 x 180). This aggregate capability of multiple drives working together
is the basis for determining the overall potential of a storage subsystem.
Since the PS Series arrays protect disk sets with RAID, the actual supported host IOPS is less than the disk IOPS
(sometimes referred to as backend IOPS). For the purpose of designing a solution or troubleshooting an existing
solution, the overall backend IOPS and host-sourced IOPS needs to be fully understood. For example, an application or
volume showing 1,000 IOPS at 50 percent reads may actually be pushing 2,500 IOPS to a RAID 50 protected set of disks
(500 writes x 4 + 500 reads).
SAN Headquarters is ideal for monitoring both the application IOPS requests as well as the actual IOPS to the disks as
depicted in Figure 8.
Applications
Data Request
Spinning Disks
CPU
Memory
Milliseconds (ms)
NanoSeconds (ns)
SSDs
Microseconds (µsecs)
Millisecond (ms) <