host ports not configured for virtual port & how many paths?

[mujzeptu]

As the subject states, I am getting this error when performing a checkhealth on my 7450.

I have verified the Brocade SAN ports have NPIV turned on and the host ports on the 7450 are all default and appear correct? We have a V400 in a similair setup and the same SAN and it does not report this issue? Latest 3.2.1 MU1 P04 is running, but cannot figure out what is causing this error?

What configuration settings do others use for your host ports? Any ideas or suggestions?

[RitonLaBevue]

what reports showsys -param ?

[mujzeptu]

System parameters from configured settings

------Parameter------ --Value--
RawSpaceAlertFC : 0
RawSpaceAlertNL : 0
RawSpaceAlertSSD : 0
RemoteSyslog : 0
RemoteSyslogHost : 0.0.0.0
SparingAlgorithm : Minimal
EventLogSize : 3M
VVRetentionTimeMax : 336 Hours
UpgradeNote :
PortFailoverEnabled : yes
AutoExportAfterReboot : yes
AllowR5OnNLDrives : no
AllowR0 : no
ThermalShutdown : yes
FailoverMatchedSet : no
SessionTimeout : 01:00:00

Is there something specific we are looking for here?

I should also mention since I did not in my original post that the esxi host settings are fine for the HBA and these same hosts have no issues to our V400 on the same fabrics - it is just the 7450 that seems to have the problem.

[RitonLaBevue]

Can you write there the result of showhost ?
It seems host ports are not well setted.
Same ports of cluster nodes to same hba: port failover.

[mujzeptu]

Hmm, here is one host example:

32 EncETF8-b10 VMware 50060B0000C28626 1:1:1
50060B0000C28624 3:1:2
50060B0000C28626 0:1:2
50060B0000C28624 2:1:1


So, the same slot and port number from different nodes must go to the same host in order to use virtual ports? What are the pros and cons of using virtual ports?

With our v400 I did this with only 2 controllers, but I wasn't aware this was required on the 7450 with 4 controllers in order to provide virtual port redundancy.

This cannot span between node 0 and 2 or node 1 and 3? It has to be between the pairs even though its a 4 node system?

[RitonLaBevue]

Ok.


What say 3PAR Best Practices. One host must see both nodes of same pair.
Then read Best Practices about Port Failover.

And the result could be:
32 EncETF8-b10 VMware 50060B0000C28626 0:1:1
50060B0000C28626 1:1:1
50060B0000C28624 0:1:2
50060B0000C28624 1:1:2

If you want your host to see the 4 nodes:
32 EncETF8-b10 VMware 50060B0000C28626 0:1:1
50060B0000C28626 1:1:1
50060B0000C28626 2:1:1
50060B0000C28626 3:1:1
50060B0000C28624 0:1:2
50060B0000C28624 1:1:2
50060B0000C28624 2:1:2
50060B0000C28624 3:1:2

[mujzeptu]

Ok great, thank you for the responses. I was under the impression that all 4 nodes could communicate for redundancy and not just a cluster pair (0 and 1, 2 and 3) so for redundancy I spread the cluster pairs across fabrics. With HP VC you cannot talk to different fabrics so looks like im going to be redoing some cabling.

Thank you for the clarification and help.

[mujzeptu]

One last question, with 3PAR being round-robin for pathing policy, I had thought esxi is only suppose to see a total of 4 active paths to a datastore LUN at a time. If I have a 2-port HBA shouldnt I just be giving it 2 paths per port for a total of 4 paths vs a total of 8 paths?

[mujzeptu]

Interesting, I re-read the vmware/3par recommendations and they mention 8 paths from a host to a datastore in it. I have always understood it as 4 as you can run into overhead, but that may be because of this:

"When configuring an HP 3PAR StoreServ Storage for use in a VMware vSphere environment, it is important to follow the recommendations for the maximum number of host server ports per HP 3PAR StoreServ Storage target ports, 128 Fibre Channel host initiators per 8Gb port and 64 iSCSI/FCoE host initiators per port."

We are actually not far off from hitting the 128 FC connections per host port as we are creeping up on 256 hosts and only have 16 FC host ports on this array.

I am curious what others have ran into or found works best? We have 2 arrays going to each esxi host and I assumed 4 paths per array was adequate, but maybe it is time to re-evaluate this?

[RitonLaBevue]

For each array you can zone some host (hypervisors ?) on a node pair and the others hosts (aix, physical intel,...) to the other node pair.
Have a look to SR port perf to split ios.