I read what I could find on this particular configuration but a lot of the information was from 2005. I am looking for current real world functionality for my configuration.
Here are some details:
Filer Configuration:
- FAS 3020 Cluster running OnTAP 7.04 (soon to be 7.23)
- Volume-per-VM, Virutal Access iSCSI RDM LUNS for each system drive within the volume
- Etherchannel Virtual interface using 2 @Gbps NICsto a redundant switch configuration (dedicated iSCSI VLAN)
Relevant ESX Configuration Info:
- Patched up 3.02 ESX Enterprise 3 node HA/DRS cluster + latest VC, VCB 1.3, latest tools on each VM
- Scripted back end hardware snapshots against RDM virtual access LUNs using VCB to quiesce LUNS
- vSwitch0 - MAC based hash load balancing using 2 @ Gbps NICs running VST on 802.1q trunks into a redundant switch configuration (VM VLANS, SC, Backup vmkernel)
- vSwitch1 - MAC based hash load balancing using 2 @ Gbps NICs one the dedicated iSCSI VLAN (vmtotion enabled, SC for iSCSI access)
- Software Initiator used and functional using single path to LUNS
Questions:
This configuration is listed as supporting iSCSI failover. I have seen a few different things mentioned insofar as OnTAP limitations on giveback that breaks failover once the NetApp cluster performs a give-back. Some people mention explicit bus resets against a given LUN on the service console back end will bring a LUN back, or that having active I/O on a LUN during give-back will cause the failover to work correctly. Information seems spotty.
- Can someone please convey the current state of affairs on this functionality and whether it should work correctly? I ask because I have found failover using the NetApp clustering against RDM luns to be inconsistent. Sometimes works, sometimes does not.
- Is there something I could be doing differently?
I have plans to break vmotion out onto its own vSwitch but I highly doubt this effects correct iSCSI failover in this configuration. Any thoughts?
