Sites - Re-Balancing iSCSI sessions?

[Paul Hutchings]

I have a multi-site cluster with a single VIP (so vsphere can use MPIO).

I noticed that so far all my vmfs volumes are using one P4500 node in one site as their gateway connection, which I believe means the other node is sat idle doing nothing?

I have the vsphere servers assigned to the site that each is in.

Would I be correct in thinking that what is happening is that the first vsphere box is in Site A so when it does an iSCSI rescan and connects to LUNs it uses the P4500 in Site A, so that's the gateway.

Then when I add another LUN if the vsphere box in Site A connects to it first, it also uses a gateway in Site A?

How can I rebalance the sessions so that both nodes in both sites are seeing equal access?

Or have I got it all wrong and it's just co-incidence/luck that of 2 vmfs LUNs the gateway connection has been through the P4500 node in Site A?

Thanks,
Paul

[Paul Hutchings]

Bump?

Seems with the Windows DSM MPIO that a similar thing is happening.

Since we're using a single subnet multi-site cluster and a dedicated SAN LAN with 10gbps fiber interconnect would it make sense to create a "dummy" site named "servers" just to assign the servers to?

In performance terms I'd sooner the servers connected to nodes in both sites - the latency isn't an issue as the sites are 1/2 a mile apart on direct fiber.

[Jay Cardin · Accepted solution]

Hi Paul,

I am seeing pretty much the same thing with my ESXi hosts but my Windows 2008R2 VMs that mount LUNs via the HP DSM do connect to each NSM (they have two connections to the NSM gateway). I have attached a photo of what the CMC is showing.

The rebalanceVIP CLIQ command doesn't split the paths for the ESXi MPIO sessions (they are always paired to the same NSM. I am still on 8.5 so maybe that will be fixed in 9.x???

I can see via MRTG that even though a host is connected to one NSM gateway, it is reading from all of the NSMs that have that LUN striped across. So I don't see the other NSMs as idle.

Check out this blog on MPIO with the P4000 : http://jpaul.me/?p=56

[Ben02]

Can someone, anyone, finally tell me if VMware, specifically vSphere 4.0, will ever utilize MPIO to create multiple gateway iSCSI connections to DIFFERENT P4000 nodes?  

This feature worked with physical Windows Servers running the MPIO DSM.  However, with ESXi, it doesn't.

vSphere shows the hosts' two NICS that I have bonded; however, the P4000 CMC maps each LUN to the same node across all three of my clustered hosts.

Please see pic attached.  I hope someone can finally walk me through setting this up or at least confirm that it's not possible.

My reason for wanting this is because with the Windows MPIO DSM, if a node failed, there was ZERO downtime.  With VMware, if the connected node goes down, I have a noticeable, several second delay that affects my production vm's.

Thanks!

Ben

[NzO]

Check your datastore configuration, the datastore must be in round robin