HPE Storage Users Group

Hi,

I stumbled across this great forum looking for best practices for 3PAR zoning and FC cabling. I guess this subject has been discussed to great length but there is a lot of inconsistency about the subject in blogs and official documentation.
We have a 3PAR 7400 4-node with 2-HBA in each node. The hosts have 2-port HBA connected to fabric A and B.

My own conclusion is as following:

1: Use Single Initiator -> Single Target zoning

2: FC cabling from each Node: port 1 -> Fabric A | port 2 -> Fabric B

3: The 2-port HBAs in a vSphere hosts should be zoned to a single node-pair (no need to connect to all nodes). The following host is then flipped to the other node-pair to balance the load (There is some inconsistency out there – some say that one shall zone to all four nodes)

4: The node ports from a host’s zoning should be mirrored across nodes to support HP 3PAR Persistent Ports.

Example:
Host1 HBA port1 -> 0:1:1 and 1:1:1 | Host1 HBA port2 -> 0:1:2 and 1:1:2
Host2 HBA port1 -> 2:1:1 and 3:1:1 | Host2 HBA port2 -> 2:1:2 and 3:1:2
Host3 HBA port1 -> 0:1:1 and 1:1:1 | Host3 HBA port2 -> 0:1:2 and 1:1:2
...

Is this correct?

/Søren Emig

Yes that all sounds about right, the only other factor I might add is they recommend a max of 32 Virtualising Hosts per array port but a max of 64 normal Hosts. So you may want to group your Virtual Hosts on one node pair and the others on the 2nd node pair. If you don't expect to reach that many hosts on the array then split them even and aim for a max of 32 per port before adding more cards. Careful not to group all the heavy hosts on one pair.

single initiator > single target is old best practice. We worked with HP a while back and they finally blessed single initiator > dual target to help cut done on zoning. They support multi targets as long as they target are the same controller pair.


We do the following

host 1 hba 1 > nodes 0:1:1 and 1:1:1
host 1 hba 2 > nodes 2:1:2 and 3:1:1

then on next host we do

host 2 hba 1 > nodes 2:1:2 and 3:1:1
host 2 hba 2 > nodes 0:1:1 and 1:1:1

We flip the primary, now since we are running ALUA this is really not necessary, but we do it anyway.

We also have set of array ports dedicated to vmware host, and others to AIX target and third set to SQL database servers.

The way you show it you are forcing all the IO form a given host to a single controller pair, while this is acceptable we prefer to have every host have the ability to let every host send IO to all controller nodes, this way if there is pressure on a given node or array port the host will send IO down a different path and possible to the other controller pair.

ailean wrote:Yes that all sounds about right, the only other factor I might add is they recommend a max of 32 Virtualising Hosts per array port but a max of 64 normal Hosts. So you may want to group your Virtual Hosts on one node pair and the others on the 2nd node pair. If you don't expect to reach that many hosts on the array then split them even and aim for a max of 32 per port before adding more cards. Careful not to group all the heavy hosts on one pair.

Thanks

In the HP 3PAR and vSphere 5 Best Practices Guide page 5 "Single initiator to multiple targets per zone (zoning by HBA). This zoning configuration is recommended for HP 3PAR StoreServ Storage"

This will surely make zoning more simple if it's true??

kluken wrote:single initiator > single target is old best practice. We worked with HP a while back and they finally blessed single initiator > dual target to help cut done on zoning. They support multi targets as long as they target are the same controller pair.


We do the following

host 1 hba 1 > nodes 0:1:1 and 1:1:1
host 1 hba 2 > nodes 2:1:2 and 3:1:1

then on next host we do

host 2 hba 1 > nodes 2:1:2 and 3:1:1
host 2 hba 2 > nodes 0:1:1 and 1:1:1

We flip the primary, now since we are running ALUA this is really not necessary, but we do it anyway.

We also have set of array ports dedicated to vmware host, and others to AIX target and third set to SQL database servers.

The way you show it you are forcing all the IO form a given host to a single controller pair, while this is acceptable we prefer to have every host have the ability to let every host send IO to all controller nodes, this way if there is pressure on a given node or array port the host will send IO down a different path and possible to the other controller pair.

Aren’t you breaking persistent port rules with this?

No the port numbers need to be the same on each node, there is a typo that last 2:1:2 is really 2:1:1

We just went 3.1.2 MU2 > MU3 and none of the hosts noticed a dropped path.

i fully agree with kluken's zoning. that is the HP recommend way as i use it as well.
uneven ports on each node to uneven fabric, even ports on each node to even fabric.
connect a host to each node of a nodepair via each fabric.

offtopic: if you need more performance, as it could be a really heavy host, then zone it to a port on all 4 nodepairs, and not a secondary set on the first pair. this might prevent upgrade issues later where node-load might block the firmware upgrade.

Thankyou all

skumflum wrote:3: The 2-port HBAs in a vSphere hosts should be zoned to a single node-pair (no need to connect to all nodes). The following host is then flipped to the other node-pair to balance the load (There is some inconsistency out there – some say that one shall zone to all four nodes)

I agree with Keith's layout as that's how I do it hear as well. 1 Host -> 4 nodes.

SAN-A Zones:
Host1 port 1, 0:1:1, 1:1:1
Host2 port 1, 2:1:1, 3:1:1

SAN-B Zones:
Host1 port 2, 2:1:2, 3:1:2
Host2 port 2, 0:1:2, 1:1:2

To make zoning easier, I have aliases for my 3PAR front end ports labeled in sets, and I rotate hosts through the sets to keep them balanced.

On SAN-A:
T800-1_4-5-1_SET1
T800-1_5-5-1_SET1
T800-1_6-5-1_SET2
T800-1_7-5-1_SET2
T800-1_4-5-3_SET3
T800-1_5-5-3_SET3
T800-1_6-5-3_SET4
T800-1_7-5-3_SET4

On SAN-B:
T800-1_4-5-2_SET4
T800-1_5-5-2_SET4
T800-1_6-5-2_SET1
T800-1_7-5-2_SET1
T800-1_4-5-4_SET2
T800-1_5-5-4_SET2
T800-1_6-5-4_SET3
T800-1_7-5-4_SET3