iSCSI, CLARiiON and vSphere – nice fix

For people with experience using CLARiiONs with iSCSI, you know it’s simple, and it’s easy.  Particularly with vSphere and FLARE 29 where we first introduced VM-awareness (vCenter API integration embedded in the platform).

But… we’re certainly not perfect.   You also likely know about a platform behavior (I think it’s fair to call it a “bug”) particularity that existed prior to FLARE 30 (the version announced today)

• when an iSCSI host tried to login to the same storage processor on the same ports (because the host had more than one NIC, and they were all on one subnet)
• the CLARiiON would assume they were the same initiator (since it dropped the source IP in the iSCSI initiator, and only looked at the IQN), and would drop the one that logged in first.
• The iSCSI initiator (and this is true of the Microsoft iSCSI initator and the vSphere initiator) would tend say “hey, I just got disconnected” and try to log back in, kicking off the other one – then repeat.
• The net symptom is very slow performance in iSCSI use cases.

…and the “pre FLARE 30” workaround

• Configure the multiple iSCSI initiator interfaces (aka the host network interfaces) on different subnets
• Configure the CLARiiON SP ports to be on those different subnets
• ensure they can’t route
• thus ensuring that each iSCSI initiator logs into one SP port and one SP port only.

I discussed this issue at length here, along with workarounds.

The downside is that you couldn’t create fully meshed network configs.

So, if its a workaround (and all platforms have them) – really the right thing would be if the CLARiiON dealt with that case better.   And.. as of FLARE 30, it does!

In FLARE 30 As shown in the diagram below, you can see that you can have a host with multiple NICs on the same subnet

 

Thanks and shout-outs go to Sheetal Kochavara and CLARiiON eng.   One thing that I think is also cool is that this is just one of many examples of the benefit we’ve reaped (and our customers benefit from) from having the CLARiiON and Celerra engineering orgs merge.   Loads of iSCSI expertise in both – put them together and you get acceleration.

Thanks for fixing this!

Comments

[Benjamin Wiechman]

Any idea when or if this fix will ever be ported to the AX4?

[Hussain]

Does this mean I can configure my AX4-5i as Active / Active controller and I can do port binding / multipathing?

Thanks,

[Chad Sakac]

@Benjamin, @Hussain - unfortunately the fix won't make it to the AX family.

[Dan Barr]

Chad, I assume this fix was included in the base Flare 31 code for VNX? Is the recommended iSCSI configuration for VNX a single-VLAN approach? Thanks.

[Chad Sakac]

@Dan - yup, it's in the VNX OE (operating environment) code - the iSCSI configuration is simpler if you use the single-VLAN approach. There are RARE cases where the multi-subnet solution can be a good choice (if you're approaching the initiator count limits - which are high).