Lefthand P4300 - MPIO with Hyper-V CSV - Problem

[Pandurang]

We have also same issue. Have you got any solution for this or limitation...

Thanks
Pandurang

[Andrew Steel]

Hi Darren,

My first thought is to chek the uplink (or trunk or whateve you want to call it) between the switches - make sure they are passing the iSCSI VLAN.

I can't see your message while replying (which is slightly annoying) - I think you said you had the NIC's bonded and ALB running. If they are patched into seperate switches then switches also need to see each other (and pass the iSCSI traffic).

If you verify that's all OK then I'll need to think a bit harder...

Cheers

[Sergio Alves]

Hi,
I was having the exactly same problem... It was the Trunk between the two iscsi Switches.
The synthoms that i had were related to communication between the servers and the storage, like pings..

Cheers!

[Andrew Steel]

Another update on the original issue:

After much playing around I settled on using 4 node windows clusters with the 4 node san. I removed the 2nd iSCSI nic and allocated it to a dedicated live migration subnet.

The problem persists that if you have too many iSCSI connections the CSV volumes fail. I have tracked it down to a problem with the persistent reservations - don't know what the solution is except to use a smaller number of iSCSI connections (less cluster nodes, less nics, no MPIO etc).

If anyone else comes across the issue and finds a solution I would be keen to know.

Cheers

[jim.hendo]

I have a similar issue with the Lefthand DSM installed.

Test Lab as follows:
2x DL360 G6
P4300 SAS Starter SAN
Server 2008 R2 Core
6x NIC

NIC 1 - Management
NIC 2 - Live Migration
NIC 3,5 - HP Team for VM's
NIC 4,6 - iSCSI MPIO

At first I install MPIO with the MS iSCSI DSM and set up multipath with Round Robin. The CSV cluster worked and the csv come online, everything was running fine.

After researching more into Lefthand SANs I found the Lefthand DSM. I installed the DSM on the first node and everything was still ok, I was able to live migrate my VMs to this node. After installing on the second node both my Quorum drive and CSV fail to start.

Thought this might be a problem with the MS DSM and Lefthand DSM both installed on the nodes and causing conflicts. So rebuilt both nodes this afternoon.

Installed OS
Installed Hyper-V Role
Installed Windows Updates
Installed PSP Pack and Teamed NICs
Installed latest Lefthand DSM and configured iSCSI Multipath
Set a 1GB LUN for Quorum and a 500GB LUN for CSV

Cluster Validation now fails because it says my shared storage does not support persistent reservations......

That's as far as I have got today, the failed cluster validation made me give up and go home... lol

So now thinking to rebuilding Nodes again and giving up on the Lefthand DSM, my SAN is mirroring over the 2nodes so I do not think I'll get much of an advantage out of using it.

Can you see any reason why the Windows DSM shouldn't be used?


Thanks

[jim.hendo]

Ok more testing...

Have setup 2x Core R2 VMs and setup the OS untill the MPIO is about to be installed. Took snapshots of the VMs.
Installed the Lefthand DSM MPIO software and setup iSCSI network.

Cluster validation fails because the storage does not support persistant reservations!!!!!

Reverted snapshots and installed the Microsoft MPIO DSM, then setup the iSCSI network.

Cluster validates and builds perfectly fine!


What gives?

http://www.microsoft.com/presspass/events/virtualization/docs/LefthandHyper-VBrief.pdf

This document from Microsoft says:

Beyond the key features mentioned above, SAN/iQ has a rich set of
additional storage management features, including:
iSCSI network load balancing (via Microsoft MPIO)

So Microsoft seem to be saying its ok to use thier MPIO DSM and not Lefthands.....

[AgdataIT]

I'm having similar issues. I have a 2 node setup as of right now with 2 connections on each node to the isolated iscsi network. The HP LH Support Team says that iscsi with the HP DSM can only support Failover. This is a serious problem in my eyes. How can you create a device that can't increase throughput with MPIO? I am now limited to 1- 1gb connection from my VM Host and Guests to their data source. This makes no sense. I attached the doc that I got from them.

I have seen someone else has played with not using the HP DSM and just sticking with the MS loaded module.

Will this work for a 2 node 08 R2 CSV Hyper-V Cluster? Any caviats?

Thanks in advance

[IainS]

AgdataITm, this is the issue I've been asking about in a recent post too and been in touch with LeftHand support.

The current 8.5 version of the DSM apparently only does failover and not load balancing across more than 1 server NIC. This is despite the manual suggesting it does do load balancing and caused me much frustration.

You can use just the Microsoft DSM and it does load balance very nicely. However it seems to only talk to one node at a time since it isn't node aware like the LeftHand DSM.

So with the Microsoft DSM you get both NICs used (assuming 2) so better peak bandwidth, however there is a noticeable reduction in write performance and iOPS for random i/o over the LH DSM. This is just with 2 LH nodes so I imagine it gets worse with more nodes.

I've reluctantly decided to go back to the LH DSM since I think we will get better performance with multiple host Hyper V nodes doing things at the same time, and database i/o etc and reluctantly accept the peak bandwidth reduction from about 190-200 MBps to about 110-115 MBps.

Apparently V9 will support load balancing and timing for it is late this year to early next (what LH support told me).

Also there are a couple of patches you may want to look into getting if you have not already. Patch 10078-02 LH support in particular was keen to get me to install (update to network drivers on LH nodes), and patch 10085-00 seems to address the issue other people in this thread are having with lots of hosts and nodes causing CSV to go offline. I stuck them both on our LH nodes yesterday and so far all seems well.