Configuring Multiple iSCSI Targets On MD3220i

As I understand it with this iSCSI san. You can only aggregate as many ports as are on each controller.

To make ESXZ use multiple paths?

Click on the host, then configuration tab, then storage. Single click a given lun, then properties, then manage paths, then choose round robin for the path selection.

If all your targets are showing up under Storage Adapters on the iSCSI adapter, this should fix it. Do some googling too, as you can tweak this for better performance some.

My knowledge of iSCSI is not that good but I am learning.

I have four vmkernel ports with one vmnic each per esxi host. All the vmkernel ports are bound to the software iscsi initiator.

Does this mean the MD3220i can have only 8 iSCSI targets (dual controllers with 4 iSCSI ports each)? Does all initiators consider the entire MD unit as a single target? If that's case the case, then how the iSCSI session/ iSCSI connection work with this unit? Here is my iscsi session info:

vmkiscsi-tool -l -E vmhba##

Session info [isid:00:02:3d:00:00:01]:
- authMethod: NONE
- dataPduInOrder: YES
- dataSequenceInOrder: YES
- defaultTime2Retain: 0
- errorRecoveryLevel: 0
- firstBurstLength: 8192
- immediateData: YES
- initialR2T: YES
- isid: 00:02:3d:00:00:01
- maxBurstLength: 65536
- maxConnections: 1
- maxOutstandingR2T: 1
- targetPortalGroupTag: 1

I have multiple of these. So it seems I have multiple sessions with one connection each. I am using Round Robin for each of my LUNs. Dell rep told me I can have only one iscsi target in the unit. Anyway I am all confused with MCS v. MPIO. All I know is MCS works at the protocol level and MPIO at the higher (host) level.

So I will just put all my esxi hosts in one group and map it to each LUNs/virtual disk.


Thanks

8 targets only 4 active. It is bacause controller ownership is per lun. Do you see 4 active under manage paths on the luns? 4 should say stand by.

You only get one initiator name if that is what your asking?

So in essence you get many iSCSI Sessions to each one connection. There are only so many iSCSI sessions supported per lun though. Each new lun gets a new set / number of iSSI sessions.

I wish a Dell rep would chime in, I am out of my league as we only have MD3000i's, but that is how it works on th MD3000i.

"8 targets only 4 active"

This is where I am lost. Are there 8 targets or 8 sessions or 8 MPIO paths? I guess I am lost in the lingo.

Dell rep tole me that I won't see the T1, T2, T3... in the runtime name of each LUN. He said the only runtime name that will show up is vmhbaXX:C0:T0:LX.
vmhbaXX = software/hardware iscsi initiator ID
C0 = This is a channel ID and will never change for ESXi4.1
T0 = Target ID
LX = LUN number assigned by the SAN unit.

The question is when will I see the T1, T2, T3, T4...?

If what the Dell rep told me is right, then I think "a" target in MD3220i or even MD3000i is the entire SAN unit. I also think each iscsi port on the controllers accepts one iscsi session for a total of 8 from each host (This would be 4 in MD3000i). The "maxConnections" value from above is 1 which means one session has exactly one connection to the SAN unit. Initially I thought a single target will be hit with multiple sessions from different hosts. That seems to be still true but the sessions are spread out on the 8 ports. There is also the Multiple Connections per Session (MCS) feature in iSCSI. MCS enables a SAN array to accept a session with multiple connections in it. Here is where the multiple targets will come in handy, I think. From what I'm reading on the web, none of the MD3xxx series units supports this feature. If you have the vmware, than your only way to loadbalancing and redundancy with SAN is via MPIO. It doesn't matter if the SAN is FC or iSCSI as MPIO is implemented at a higher level than the SAN protocols. Now I could be completely wrong with all this. I also wish Dell reps would chime in and help us. Thanks for your help.

Hi p3pp3rjack and johnadco,

I've asked PG Engineering for clarification. Awaiting their response.

Pretty darn sure it is each lun that decides the concurrent number of iSCSI sessions. The amaunt of "sessions" is not determined by the amount of connections.

We are using vCenter 4.1 and have multiple ESX servers. We just went through an ordeal with the MD3220i and MPIO/multi-pathing. I just posted a thread on that subject asking expert advice for SMCLI information.

We spent a considerable amount of time with VMware and Dell Storage logged into our system. What we came away with is that VMware will identify a path for every port on the controller to every LUN. However, the MD controller will in fact direct VMware to use a single "optimal" path for each disk group. So while the paths are in Vmware you cannot use them. We tried with Round-Robin and fixed pathing. We had 24 paths in each ESX server but only one would be active, across all servers. Usually the same path on all servers. That is because we had one disk group on that controller with multiple LUNs. LUNs do not appear to be relevant to the pathing for the MD3xxxi. The controller assigns optimal path based upon disk group.

VMware Tech said the T1 and C1 labels should not be referenced, those are internal values on ESX that can change and are not necessarily tied to consistent MD32 ports/controllers, to verify your paths look at the full IQN and see the IP of the port on the controller to ensure multi-pathing. Although we had different T1,T2... values, when we checked the IP it turned out all of our servers were using just a single port on each controller - 6 ports were unused.

To correct we had to break out the MD32's into smaller and more disk groups. Now, with ESX set for MRU we have multiple paths.

The MD3220i kind of turns the mantra of "more LUNS is better" on its head.

I was able to get MPIO working to a 3220i with 2 Raid 10's 1.64 TB on each side. It works fine when 1 R610 server is running it to the SAN. I enabled MPIO on 2 more R610's and Hosts would randomly freeze. I am in the process of trying to figure out what is causing that and eliminating the issue. This does go over 2 dell E class switches, all on the same subnet with no Vlans. I may try it with procurves and even separating them into VLAN's.

I may just abandon MPIO all together and just give each 610 its own port on each side of the raid controller. One problem I have run into is removing the separate VMKernels. If I remove them from the GUI I get errors when I tail the logs.

Nov 12 18:53:12 iscsid: Could not bind connection 0 to vmk4 (Error=22,Invalid argument)
Nov 12 18:53:12 iscsid: cannot make a connection to 192.168.130.108:3260 (22,Invalid argument)
Nov 12 18:53:12 iscsid: Could not bind connection 0 to vmk4 (Error=22,Invalid argument)
Nov 12 18:53:12 iscsid: cannot make a connection to 192.168.130.103:3260 (22,Invalid argument)

This is very frustrating all around, as support from both Dell and VMware engineers is sketchy at best.

If anyone has some suggestions, I am all ears.