Administrator Guide

the maximum number of random IOPS that can be sustained. EqualLogic customer support or your channel partner can help size
storage congurations for specic workloads.
Also, review the latency on your servers. If the storage does not show a high latency but the server does, the source of the problem
might be the server or network infrastructure. Consult your operating system, server, or switch vendor for appropriate actions to
take.
Estimated I/O Load
Based on latencies, IOPS, hardware, and the RAID conguration, the estimated I/O load is relative to the theoretical maximum
capability of the group, pool, or member. Because the load value is an estimate, use it only as a general indicator. The I/O load can be:
• Low — Minimal load. Latencies are low.
• Medium — Typical load. Usually, either member throughput is greater than 1MB/second or IOPS are greater than 50, and
average latencies are above 20 ms. If this load is sustained, you should monitor the group carefully.
• High — Load that is approaching the theoretical maximum capability. Usually, member throughput will be greater than 1MB/
second or IOPS will be greater than 50, and average latencies will be above 50 ms. If this load is sustained, you should investigate
further.
• Unknown — Indicates that SAN Headquarters cannot currently calculate the I/O load because of invalid SNMP counters or
because of the group workload.
Average IOPS
IOPS are a good way to measure overall I/O activity and how much work is being done, especially if you also consider the average
I/O size. However, IOPS do not indicate whether the storage system is overloaded or operating within its limits.
Average I/O Size
In general, the average size of an I/O operation is 16KB or less. The larger the I/O operation size, the longer it takes to process,
which might aect latencies. Also, large I/O operations usually reduce the total number of IOPS.
I/O size can be useful in understanding workload characteristics, especially when measured at the volume level.
Network Load
The network load is the percentage of the theoretical maximum network bandwidth that is being used for sending I/O or receiving
I/O, whichever has the highest value. The theoretical maximum bandwidth is based on the negotiated link speed for all active
network interfaces. The network load percentage provides a quick measure of network activity. Network ports are typically full-
duplex.
The network is rarely a bottleneck in a SAN. Usually, network bandwidth is underutilized, especially with random I/O workloads.
However, bandwidth might become fully utilized during highly sequential operations, such as a backup; in most cases, fully utilized
bandwith does not indicate a problem with the system.
Using all the available bandwidth on one or more member Ethernet interfaces generates an alert.
Network Rate
In general, the network rate should be 100 percent to 200 percent of the I/O (iSCSI) trac. A network rate that is signicantly more
than 200 percent might indicate a problem.
Queue Depth
SAN Headquarters displays the queue depth (average number of outstanding I/O operations at the start of each incoming I/O
operation) for each disk drive (raw I/O), volumes (only iSCSI trac), groups, and pools. A queue depth of zero indicates no I/O
activity. High or sustained queue depths might indicate that the group is under a high load.
NOTE: A group must be running PS Series rmware version 5.0 or a later version to display iSCSI queue depth for a
volume.
About Monitoring
335