Database Reference
In-Depth Information
value and KAVG in ESXTOP. Recommended thresholds for average and maximum
values can be found in Chapter 10 .
The LUN queue depth isn't the only value that you may need to modify in order to
increase performance from your data store. The LUN queue setting goes hand in hand
with the VMware vSphere advanced parameter Disk.SchedNumReqOutstanding
(DSNRO). DSNRO is used to control the queue maximum depth per VM when there are
multiple VMs per data store. The goal of this setting is to ensure fairness of IO access
between different VMs. When there is only one VM per VMFS data store, the LUN
queue depth will always be used. In vSphere, Disk.SchedNumReqOutstanding is a
global value up until vSphere 5.5. In vSphere 5.5, Disk.SchedNumReqOutstanding is
specified on a per-device basis. This setting is modified dynamically, as is the LUN
queue depth when Storage IO Control is enabled on a data store with multiple VMs that
is experiencing performance constraints.
Tip
You can set the per-device number of requests outstanding in vSphere 5.5 by
using the command
esxcli storage core device set -d naa.xxx --sched-num-req-outstanding=<value>
where naa.xxx is the device name and <value> is a value from 1 to 256.
To list the storage devices on the system, use the following command:
esxcli storage core device list
By specifying the -d naa.xx option, you can confirm the setting has been changed
as you expected. Also see VMware KB 1268 for further information.
Figure 6.23 shows the different queues at each level of the vSphere storage architecture.
The two values that are usually worth monitoring as a vSphere admin are the AQLEN
and the DQLEN. DQLEN can be adjusted up or down, depending on your requirements.
For high-IO SQL Server systems where PVSCSI is used on VMDKs, we suggest you set
the DQLEN to 64 as a starting point, while taking into account our previous
recommendations when modifying queue depths.
 
Search WWH ::




Custom Search