Working with Windows Server 2003 iSCSI clusters

Microsoft now (for awhile) supports iSCSI for failover cluster storage (more info here and here), that doesn't mean however that everything works as expected.

Here is a collection of helpful tips that I gathered in order to successfully get iSCSI working on a Windows Server 2003 cluster. A glimmer of hope is that Windows Server 2008 does not require the registry changes or software installation, it does seem to work properly out of the box.

• Install this update: http://Support.microsoft.com/kb/932755 (some HP servers have a prerequisite driver upgrade)
• When creating disk bindings on the cluster, create the binding while the resources are on the opposite node otherwise on reboot the server will to bind the nodes to the letters when the drives are not available.
• If using dual ported or dual HBAs:
• Each target must have two entries in the target portals, one for each HBA port.
• Create persistent multipathed enabled connections from each HBA port to each lun.
• If iSCSI HBAS are used:
• Ensure the Microsoft iSCSI software initiator was not installed and that the Microsoft iSCSI client was version 2.06 (or higher). (this is done with a custom install of the iSCSI initiator)
• Disable the HBA’s NIC devices in device manager to prevent the NIC driver from being loaded.
• Verify the HBA firmware and driver were up to date based on manufacturer recommendations.

Set these registry keys:

This disables the boot time firewall that prevents the cluster from allowing iSCSI connections to the quorum while the servers is starting

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\IpNat\Parameters]

"DisableBootTimeSecurity"=dword: 00000001

This sets persistent SCSI reservations so that each node can be uniquely identified on the cluster to handle resource failovers.

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\msiscdsm\PersistentReservation]

"UsePersistentReservation"=dword:00000001

"PersistentReservationKey"=hex:aa,bb,cc,cc,bb,aa,00,01

Note: The last two hex digits should be different on each cluster node. Node 1 would be 01, node 2 would be 02, and so on)

This sets a higher disk timeout to overcome possible iSCSI latency issues.

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Disk]

"TimeOutValue"=dword:0000001e