HP StorageWorks SAN Virtualization Services Platform Best Practices Guide (5697-0935, May 2011)

Note that ports 0 and 2 (quad 1 front-side ports) have the same count of 'Inits' (initiator remote
ports) and that ports 1 and 3 (quad 1 back-side ports) have the same count of 'Targs' (target remote
ports). The same is true for quad 2. The other DPM in this DPM group displays the same exact
values. This indicates a symmetrical configuration. Also note that the 'Flags' virtual disks all have
values that end in 3f9. This indicates that all the ports are up and in a good state. Values that end
in three digits other than 3f9 may indicate that the port is in a bad state, and you should verify
the setting.
The show ports command on the DPM displays a port in a Ready state if the DPM does not
detect any back-end LUs. For safety, the DPM will turn off the target ports.
Num State Cap. Type(conf/act) Speed Mtu
0 READY TARG FABRIC/UNSET 4G/UND 2048
1 ESTABLISHED INIT FABRIC/FABRIC NEG/4G 2048
2 READY TARG FABRIC/UNSET NEG/UND 2048
3 ESTABLISHED INIT FABRIC/FABRIC NEG/4G 2048
Monitoring for performance
Most of the components in the SAN have best practices or performance metrics that may be
compared with the performance found using the tools described below. This guide and SVSP white
papers use the term Safe IOPS to define performance that is predictable across single failures. In
general, SVSP I/O average latencies over 30 ms should be avoided and are indicative of problems
with the configuration.
To monitor SVSP performance:
• Monitor the system with the VSM GUI regularly.
• Monitor the application server-to-array throughput using the Fibre Channel switch vendor
performance tools.
• Monitor the internal VSM data moving performance using a tool for collecting performance
data like Windows Performance Monitor, which is described in the HP StorageWorks SAN
Virtualization Services Platform Administrator Guide.
• Monitor performance at the array using array-supplied tools. Ensure that the array is operating
according to the manufacturer's best practices. The following are signs of an overloaded
array:
◦ Average latency above 30 ms is an overloaded array (average latency above 10 ms is
a heavy load).
◦ Large numbers (over 100/hour) of "Queue full" SCSI responses from arrays.
• Monitor cross-sectional bandwidth. This is the sum of the bandwidth of all links between the
servers and the DPMs, or the DPMs and the arrays.
• Monitor DPM performance. See the "Monitoring DPM performance" section within the
"Monitoring the SVSP domain" chapter in the HP StorageWorks SAN Virtualization Services
Platform Administrator Guide for monitoring with the DPM Management GUI.
Array-specific monitoring
Consult the array model documentation for non-EVA monitoring information. The EVA monitors
standard Fibre Channel counters; they are only displayed when a count exceeds a threshold. The
count is checked every 60 seconds, and if any count exceeds the event trigger, an event is logged.
EVA arrays spread the load across many spindles; other arrays can present LUNs to SVSP that are
configured using very few spindles. This may lead to performance issues. See “Array configuration”
(page 18) for more information.
58 Monitoring the SVSP environment