HPE Storage Users Group

In working through some of our Performance issues HP had us setup the zoning on our Fibre switches such that each of our VMWARE hosts each sees only 4 paths to the storage. We have a dual-node 7400 which with all ports going we should 8 paths but we set it up how HP had requested which actually removed paths from the host to storage and we saw a performance decrease. How do most do their zoning? Do they do it such that the VMware host can see all paths.... Did a quick search for some sort of guide but didn't see anything specific to this right off the bat.

Thanks
Mike

So this is about fan out ratios... our standard is for each HBA port, we zone to 2 storage ports. So basically all of our servers have 2 HBA ports, which nets to 4 storage ports per host.

Too much "fanning out" has been identified as a bad thing in the past when we were a EMC Powerpath shop, so we just use the 1 HBA to 2 storage ports now as a best practice for everything.

More specifically, in your case with a dual node 7400, I would zone each HBA to one port on each of your nodes, and evenly spread them out. If/When you get 4 nodes, the next step would be zoning each HBA to a node pair, so that 1 host with 2 ports would see all 4 nodes.

We have 2 node-pairs currently and actually have an add-on fibre-channel card per node pair and an ISCSI card per node pair. Apparently, we originally did not had the zoning done not per HP best practices. The way it was before we were using I think 4 of the onboard ports and two of the ports on the add-on cards. So, we had ESX setup with 6 paths to storage. We reconfigured things per what HP was telling us which pretty much lines up with what you are saying. So, currently we have it so that each ESXi SAN port sees 2 3Par ports, 2 ports per host so 4 paths to storage. There are jobs that run during the night that after this change we saw about a 30% decrease in performance because we went down from 6 to 4 paths. My thoughts were to zone it back so that we had 4 paths per VMware Host port with a total of 8 paths to get back to where we were and maybe even gain some. Because there is a definitely a direct correlation in the loss of performance we saw when we dropped the number of paths.

Are any of the node ports flooded during the slow period? For production ESX we've used the 8 paths (the max supported I believe) for 4 node arrays as we have large beefy ESX servers. You are limited to a recommended 32 ESX hosts per port though so depending on your scale might want to only do that for clusters running high bandwidth apps.

Hi.

Was the zone configuration the ONLY change done?
Are the ports saturated?

The performance degration might indeed be related to zone and path configuration, but I would also look for other possible reasons.

Yizhar

It should also be pointed out to elaborate what Richard said that when you zone to 2 storage ports they should be on the same controller pair. HP originally wanted one to one mapping of FC zones, they now allow you to zone a HBA port to a pair of storage ports, but they need to be on the same controller pair. This helps cut down on the number of FC zones. Also in 3.1.2 with persistent port when you take a controller down for repair, upgrade or whatever the attached hosts never really see it as the port fails over to the other node in the controller pair. Obviously this applies to arrays with 4 or 8 controllers only.

Hi
you should look at 3par best practices guide
1) Best practice: A host should be zoned with a minimum of two ports from the two nodes of the same pair. In addition, the ports from a host’s zoning should be mirrored across nodes. This is to support HP 3PAR Persistent Ports.
• Example from a single host: 0:2:1 and 1:21
Hosts do not need to be connected to all nodes because of the way the volumes are spread on all the nodes. A connection to two nodes is enough.
2) Best practice: No more than 64 initiators are supported per front-end/host port.
When using hypervisors, avoid connecting more than 16 initiators per 4 Gb/s port or more than 32 initiators per 8 Gb/s port.
Regards

We actually use dual ported HBAs on our vmware hosts and one port is zoned to nodes 0 and 1 and the other port to nodes 2 and 3 and then we flip that for each group of server to balance any load, although all our hosts a ALUA.

@giladzzz
It is some kind of a strange about what you are saying regarding to "3PAR best practices" and I'll explain why. We got some kind of a performance assessment on one of our V800 3.1.2MU2 and among other things 3PAR guys told us there were the following words:"Host balance is wrong, hosts should be in different node-pairs. This is mandatory and stated in the 3PAR best practices. HOSTS should not be in the same node-pair but in different node-pairs".
I really don't know what to believe about this "3PAR best practices". I also used zoning the same way you said but now I don't know what to say. Or maybe I didn't well understand and the 2 ways of doing zoning are not quite different.

thanks,
mugurs

@Mugurs - If you copy/paste the full paragraph from the results of your performance assessment we can help break down what they were referring to in your situation... however, with out a doubt, it is BEST practice to zone a single host to both nodes in a pair, if you want 2 paths, it goes to 1 node pair... if you want 4 paths, and have 4 nodes, you should use 2 node pairs (4 nodes) to keep all 4 nodes evenly balanced.

mugurs wrote:@giladzzz
We got some kind of a performance assessment on one of our V800 3.1.2MU2 and among other things 3PAR guys told us there were the following words:"Host balance is wrong, hosts should be in different node-pairs. This is mandatory and stated in the 3PAR best practices. HOSTS should not be in the same node-pair but in different node-pairs".