Multiple VIPs on VSA cluster

[DodiB]

Hi Guys,

Is it supported to have multiple VIPs on a VSA cluster and if so is there any potential performance or resilience gain? I've got a 2 VSA cluster with a FOM on a third ESX host and when I drop the VSA holding the VIP VMware HA doesn't restart the VMs running on the iSCSI datastore as the remaing VSA in the cluster takes 15 or 20 seconds to claim the VIP, and the volume goes offline because of this while HA is trying to bring up the VMs.
So I'm wondering if I create multiple VIPs will this help with resilience while also doing an element of multipathing?

Cheers
DB

[Gauche]

It's possible to put multiple VIPs on a SAN/iQ cluster but for different reasons then you ask. It requires at the least that the nodes be actually running in different subnets. So you can't have two VIPs in the same subnet.

But what you describe means that something is mis-configured. VSA is designed, tested, and certified with VMware to do exactly what you are trying. Do you have support? I'd suggest calling to see what's up with your failover not working right.

[DodiB]

Thanks, I think I suspected as much from the docs re the VIP but wasn't sure.
With regards to the VIP timeout, this is an eval of the enviroment at the moment so don't have support. Am going to test this again though as the VSA I dropped has now developed a problem, storage server not ready showing on it for a while and I tried to repair it but it had a bit of a fit because I never set the 2-way volume it was part of the redundancy for to 0-way it seems. Also possible that the network here s causing a problem, Will post back when I've retested if I find anything interesting.

Cheers
DB

[Gauche]

You should not be changing the replication level lower to repair the node, just repair the node. But I'd bank on a network issue, it always is.

[DodiB]

I have a follow up related question on the use of multiple VIPs. If I add a second subnet to my config, a second vNIC to each VSA (not sure this is supported?), and a second vmkernel port for iSCSI to each ESX host. Then configure a second VIP on the SANiQ cluster, does this give me multipathing to SANiQ volumes? And is it supported?
So to re-state that I would propose to have two paths for my iSCSI vmhba:
1. a. 10.0.0.1 - vmk_iscsi_1 - first ESX iSCSI vmkernel port
b. 10.0.0.10 - vsa_nic1 - first VSA nic
c. 10.0.0.20 - SC_net_1 - first ESX COS port
d. 10.0.0.30 - VIP_1 - first VIP
2. a. 10.1.0.1 - vmk_iscsi_1 - first ESX iSCSI vmkernel port
b. 10.1.0.10 - vsa_nic1 - first VSA nic
c. 10.1.0.20 - SC_net_1 - first ESX COS port
d. 10.1.0.30 - VIP_1 - first VIP

I then add the two VIPs as the target discovery addresses on the ESX host and this hopefully gives me two paths to each volume? My interest here is in pushing redundancy and performance as much as possible as with two paths I can then also be certain that multiple 1gbps uplinks can be used by my ESX iSCSI initiator.


Cheers
DB

[DodiB]

I'm still having some problems with failover at the cluster level. I've just tested again and it took about 1 minute for volume access to be restored to the esx hosts after I took one of them down by pulling the power. The VSA that was downed by this was running the VIP. I appreciate that some of thr 1 minute for volumes to recover at the ESX side is vmware timeout but not all of it and not even most of it I think. What is the expected failover time for a VIP usually?

Cheers
DB

[Jess1313]

Hello Dodib, 

Yes dual subnet/dual VIP multi-site clsuters are supported on LeftHand OS, however per the P4000 Multi-Site Config Guide:

http://h20000.www2.hp.com/bizsupport/TechSupport/CoreRedirect.jsp?redirectReason=DocIndexPDF&prodSeriesId=4118659&targetPage=http%3A%2F%2Fbizsupport1.austin.hp.com%2Fbc%2Fdocs%2Fsupport%2FSupportManual%2Fc01727773%2Fc01727773.pdf

When using VMWare ESX 4 or higher and its native MPIO in a Multi-Site SAN, you cannot configure
more than one subnet and VIP. Multiple paths cannot be routed across subnets by the ESX/ESXi 4
initiator. With VMware ESX, the preferred configuration is a single subnet and with ESX servers
assigned to sites in SAN/iQ.

so assuming your not using ESX3 it is unsupported for ESX.