User guide

Technical white paper | HP Enterprise Virtual Array Storage and VMware vSphere 4.x and 5.x configuration best practices

Before enabling adaptive queuing, HP highly recommends examining your environment to determine the root cause of

transient or permanent I/O congestion. For well-understood, transient conditions, adaptive queuing may help you

accommodate these transients at a small performance cost. However, to address more permanent I/O congestion, HP

offers the following suggestions:

• Using VMware disk shares

• Using VMware Storage IO Control (SIOC)

• Re-evaluating the overall ability of the storage configuration to accommodate the required workload – for example, non-

vSphere hosts may have been added to a well-tuned, balanced SAN, placing additional stress on the shared storage; such

configuration changes must be investigated and addressed before you throttle back the vSphere hosts.

If you must enable adaptive queuing, HP recommends the following values for HP EVA Storage homogeneous

environments:

• QFullSampleSize = 32

• QFullThreshold = 4

Note the recommendation for QFullThreshold is changed from eight (previous recommendation) to four. This will work just

as well with EVA and also make EVA array coexistence with 3PAR configurations much more efficient.

For environments where the ESX servers are connected to both HP 3PAR Storage and HP EVA Storage, then the following

values are recommended:

• QFullSampleSize = 32

• QFullThreshold = 4

Best practices for adaptive queuing

• Rather than enabling adaptive queuing, determine the root cause of the I/O congestion.

• If you decide to use adaptive queuing in an homogenous HP EVA Storage environment or an EVA/3PAR heterogeneous

environment, set the values as follows: QFullSampleSize = 32 and QFullThreshold = 4.

In ESXi releases earlier than ESXi 5.1, QFullSampleSize and QFullThreshold are global settings. The value set will apply to all

devices accessible to the ESX server. In ESXi 5.1, because of different recommended optimal values for various arrays, these

parameter can now be set more granularly on a per device level.

Run the following esxcli command:

esxcli storage core device set --device device_name --queue-full-threshold Q --

queue-full-sample-size S

Where S is the value for QFullSampleSize and

Q is the value for QFullThreshold

These settings are persistent across reboots and you can confirm their value by access the specific device properties with

the command:

esxcli storage core device list --device device

Using the paravirtualized virtual SCSI driver

First available in vSphere 4, the VMware paravirtualized virtual SCSI (pvSCSI) driver is installed with the VMware Tools. This

guest OS driver communicates directly with VMware Virtual Machine Monitor, helping to increase throughput while reducing

latency.

In testing, the pvSCSI driver has been shown to increase the performance of Vdisks compared to standard virtual SCSI

adapters such as LSILogic and BusLogic.

When using Iometer, for example, there was a performance improvement of between 10% – 40%, depending on the

workload used. The tested I/O block sizes were 4 KB, 8 KB, and 64 KB for sequential and random I/O workloads.

Best practices for improving the performance of VMs with I/O-intensive workloads

• Consider using the pvSCSI driver with the VM’s data logical units, which can enhance performance by 10% – 40%,

depending on the particular workload used.