HPE Storage Users Group

Just been reading a best practice guide that says for zoning you should all odd ports to 1 fabric and all even to another fabric. Which i agree with.

But they say that if you are zoning to to port 0:2:1 then you should also zone 1:2:1 and you shouldn't use one form the other node pair, i.e. 2:2:1

can anyone explain why you wouldn't do this? as i though the zoning would be safer to have 1 port going to 1 node pair, and 1 port going to the other node pair?

as the problem with having both links in to 1 node pair means if you have a power fault on that drawer, you lose both nodes.

Please advise

Originally they wanted a one initiator to one target zoning, but they have relaxed that to one initiator to 2 targets as long as the 2 targets are in the same node pair and for persistent ports they should be the same port, so you are correct that 0:2:1 and 1:2:1 can be in the same zone as the initiator and you should NOT add in 2:2:1, rather if you create a separate zone for the other initiator to that node par. This helps you balance out IO and also avoid hitting any limits for the number of initiators per target.

So my suggestion would be zone all HBA A ports to say 0:2:1 and 1:2:1 and all HBA B ports to 2:2:1 and 3:2:1 with round robin you are evenly distributing the IO load and you have good redundancy without worrying about the limits of initiators per 3par target port, which I think is something like 64 right now.

thanks for the quick reply.

i understand your idea, but i thought all odds would have been fabric a and evens would have been fabric b.

so i would have thought it would be better to be 0:2:1, 1:2:1, 2:2:2 and 3:2:2

so what harm would it cause having it zone up like 0:2:1, 2:2:1?

also we have some hypervisors that 8 paths like the following 0:2:1, 0:2:2, 0:2:3, 0:2:4, 1:2:1, 1:2:2, 1:2:3 and 1:2:4 should we have 8, and does it do any harm?

If you hava a four node system, nodes 0 and 1 mirror their cache via a direct (backplane) connection, and node 2 and 3 do so. Between the node pairs, there is "only" a cable connection.

If you zone a host to just two of your four nodes, HP advice is "use two nodes that mirror their cache". If you don't do this, you'll encounter no direct problems, but loose some % performane (I guess). We DO have some native windows-hosts that are zoned to "wrong" node pairs and did not encounter any problems so far.

We zoned our esx-host to the ports 0:1:1, 1:1:2, 2:1:1, 3:1:2 (port group A) or 1:1:1, 0:1:2, 3:1:1, 2:1:2 (port group B). With this setup, every node-pair is reachable via both fabrics for every host. 0:2:1, 1:2:1, 2:2:2 and 3:2:2 (as you suggested) makes node pair 0-1 only available on fabric <uneven> and node pair 1-2 only available on fabric <even>.

_________________
When all else fails, read the instructions.

These is nothing wrong with more paths to a point. As long as you fit within the 3par specs per host port. We were too close to the host port limits so we break it up.

If the hypervisor is VMWare be aware that there is a limit of 1024 paths per host so if you do 8 paths per volume you are limited to 128 datastores and/or RDM's per host.

thanks for the replies guys.

so you think the only reason say to have 1 fabric zoned to 1 node pair is just for cache performance if 1 node was to fail?

3PAR best practice: http://www8.hp.com/h20195/v2/GetDocumen ... A4-4524ENW

Have a look page 9 & 10.

Then you will zone your 2 ports ESX like this:
port 0 : 0:1:1 / 1:1:1
port 1 : 0:1:2 / 1:1:2

Here we proceed like this :
port 0 : 0:1:2 / 1:1:2
port 1 : 0:1:1 / 1:1:1
even to even and odd to odd.

On 7XX0, slot 1 (internal) is prefered for front end connexions. slot 2 is used for front end and RCFC.

Allways keep in mind Port Persistent.

so for port persistence, do you have to both 0:1:1 and 1:1:1 in the same zone with the port of the host?

as i think we have 1 zone per port, so for instance a host with 1 zone going to port 0:1:1 and it will also have a second zone from the same host port but going to 1:1:1

Does this mean in this configuration we wouldn't be using persistent ports?

To benefit from port persistence you need to zone 0:1:1 and 1:1:1 to same host port. No need to be in the same zone.