PowerPath on a virtual machine if I have iScsi initiators?

Hi Gabriel,

The only supported configuration of PowerPath for Windows in a VM (where Windows is running in a VM) is via iSCSI.

So, if you want to install PP in a cluster setup, it is recommended to have all cluster nodes equally configured.

so I am running PowerPath in my ESX cluster and I need to install PowerPath for Windows inside my existing Microsoft Cluster install, where my iScsi initiators are running? Correct!?

last year, I believe, we had 100% CPU usage when we tried that, so upon calling EMC support we were told that PP doesn't go on the VM, but perhpas they misunderstood our configuration?!?

We've been testing SQL Cluster running on these 2 VMs (cluster nodes) and we have been seeing I/O latencies, so we started wondering and looking around, switches look clean, VNX looks clean, VMWare confirmed yesterday that our vDS look clean, that's when I started this thread.

Hi Gabriel - yes please re-open the case with customer support and ask that it be escalated to level 2 support.

It sounds like your configuration is supported but it's an atypical scenario so support may have misunderstood the first time.

If you still run into issues please contact me directly.

Bob Lonadier

PowerPath Product Manager

robert.lonadier@emc.com

Hello Gabriel,

I think the important part to make sure is understood in your configuration is how you are connecting to your Cluster storage.

Normally, in a VMware environment, you are not required to install PP in a Windows VM (or another OS) because you might already have PP installed in your ESXi hypervisor. This is because most if not all storage is coming directly to the ESXi which then is used by the VMs as VMDKs. So, all of the process and IO will be handled by the ESXi host thus needing PP (in the ESXi host) to optimize the IO via all paths.

However, if the Windows VM is accessing the Cluster storage via iSCSI, completely bypassing the ESXi hypervisor (this is technically speaking although we know that the vNICs are still going through ESXi), that means that the storage is NOT seen by the ESXi host at all. In this situation, yes, PP can be installed in the VM simply because you are connecting to your storage via iSCSI.

Must of the problems that I see with iSCSI configuration (whether physical or virtual server) is that is it not setup correctly. I highly recommend taking a look at the Host Connectivity Guide and take a look at all of the iSCSI Attached Environment section (page 38). For configuration see starting in page 54 to make sure it is done correctly (applies for 2012 R2).

https://support.emc.com/docu5134_Host_Connectivity_Guide_for_Windows.pdf?language=en_US

when I am running SQL and do load test, I now have powerpath on both nodes now, I have 2 iScsi VLANs (10.131.2.0/24 and 10.131.200.0/24) so I assigned 1 IP per VLAN, 2 virtual NICs on each node.

Is powerpath suppose to "drive" 100% of the iScsi traffic via only one of them (the one with the best path)?? or is it supposed to go 50/50?

I am seeing data going thru both of them at the same rate (circa, tomorrow I will have more precise values from perfmon, esxtop, extrahop, mrtg, etc..).

I do NOT have LACP, I am also wondering if I should add more physical interfaces of the host into the iScsi VDS or not. Shoudl I also add more virtual NIC to each node, like 2 + 2 for better throughtput? Would that help powerpath to make better path choices?

Thank you

When PowerPath is installed with a license it will load balance the I/O through both NICs. Isn't that what you would want?

Regards, Michel.

Hi Gabriel,

One of the benefits of PowerPath is that it will Load Balance the IO via all the paths. This effectively will utilize all paths equally which is why you see basically the same rate via all the paths.

It is completely up to you if you want to add more NICs. This will allow PP to use more paths for IO.

do I need PP on the hosts and the guests at the same time for this??

I've been migrating data from a LUN to another LUN, using this virtual machine with the virtual iScsi initiators on and I don't think I am getting good results. My average speed of transfer is 30MB/s, and I think that's just 25% of a single Gigabit interface.

If we think that (since I am copying from LUN to LUN) I am using the iScsi connections to read and write, I would at least expect a 50% speed (50 to read and 50 to write simultaneously).

Am I wrong? where should I look at now?

The thing is that I've been troubled by this for a bit now and all the vendors support (EMC, Cisco for switch, VMWare) tell me to ask each other and nobody helps me much. while this is the place I've found be most useful so far, so here I am back again

Hello!

Do you mind explaining a little more what you are trying to accomplish?

The important part to remember is how are you presenting your storage to the VM. That will determine where PP should be installed. Basically if you are presenting your storage to the hypervisor (i.e ESXi) then PP can be installed there (you then can present a vmdk or as rdm to the guest). If you are presenting storage to the VM directly (i.e. iSCSI), which means the hypervisor has no knowledge of that storage, then PP can be installed there.

The way i understand it, sounds like you are migrating LUN to LUN inside of your VM via iSCSI. What tool are you using to migrate? Or is it just normal copy (i.e drag and drop or copy/paste)?

that's correct, normal copy, windows explorer, both drives are in the same cluster resource, so they are attached at that moment with the same iscsi iniators (virtual nics).

Currently I have both PP on ESXi and VMs (and I think it's correct), now that you asked me how I am migrating LUNs, I just thought of doing SAN copy. Is that what you were going to tell me? Do I need a license for that ?

Thanks for the info. Since you are connecting via iSCSI inside the VMs, PP in the ESXi is NOT needed. Unless, of course, you are also connecting via Fiber for other purposes (i.e. VMs via RDM).

Now, for the VM migration. Something that is important to remember, assuming iSCSI was configured as per Host Connectivity Guide, if you are just doing a normal copy, windows explorer, the OS will treat it as a sequential copy and that usually means a single path will be used and the copy is one frame at a time. This potentially is the reason why you are only getting the speeds you got. Have you looked at using PPME? If you have 5.7.x and up, the normal license will allow you to use it.

Unfortunately, i am not familiar with SAN Copy, so i am not able to properly comment on that.

what is your source and targer array ?

trying to migrate from sql 2008 to 2014 MS Cluster, the Cluster nodes are moving from 2008 to 2012 R2, there is a 1GB/s firewall in between of them, so I didn't want to copy the data thru it.

I attached the new drives to the old cluster and did a windows copy, ~ 500GB, I was looking for the faster way to do that.

VNX5300 series with 1Gb iScsi switches

Since we are in the PP group, I will say PPME would be the fastest way, but it will leverage the network which syncing.