3.5.1 Matrix Server Administration Guide

Chapter 18: SAN Maintenance 304

Host Bus Adapters (HBAs)

Reduce the HBA Queue Depth

The HBA queue depth is the maximum number of outstanding I/O

requests that the HBA can hold while it is waiting for responses from the

LUNs on the storage array. If the HBA is sending more I/O requests than

the array queue can handle, you may see I/O performance problems such

as the following:

• “Recovered I/O” error events in /var/log/messages

• Sluggish I/O in Await or SVCTM as measured in iostat -x 5

• Reports of “DLM lock stalls” for locks over filesystem I/O

You may be able to eliminate these problems by changing or adapting

your workload. For example, you might move jobs from a heavily loaded

node to a less busy node or reschedule certain activities.

If you are unable to change the workload, try reducing the HBA queue

depth, as described below.

NOTE: This procedure should be used only if Matrix Server controls the

loading of the HBA driver. If you are using third-party MPIO

software such as EMC PowerPath, HDLM, or IBM RDAC in which

the HBA driver load is controlled by the initrd image, do not use

this procedure. Instead, changes to the HBA queue depth should

be made only under the advice and direction of the third-party

MPIO vendor.

Determine a Value for the HBA Queue Depth

Consult the hardware vendor for your array to determine the number of

supported SCSI command tags per port/LUN. Then divide that number

by the number of servers in the cluster. For example, if the storage queue

depth is 32 and there are eight servers in the cluster, you would use 4 as

the HBA queue depth.

If you are unable to determine the queue depth for the array, try setting

the HBA queue depth to 4.