Redundancy Failure testing EVA 4400 controllers

[linuxbass]

Testing EVA 4400 redundancy.

To test controller redundancy both shelve bus device ports were disconnected from controller 2.

1st disconnect gave an alert in CV as expected. Disconnecting the 2nd and the array failed completely including access to CV.

Both device ports disconnected from controller 2 access to storage and Command View went offline. Internal mgmtmod CV could not "take over" control from the management server CV.

Reconnecting the cables did not help either as they port lights did not illuminate and it appeared that the controller went completely offline.

This seems a simple and reliable test for controller failure however the Array basically fell over.

Has anybody tried such a test with the EVA 4400?

I am worried that this shows a huge vulnerability with the EVA 4400 equipment. In my experience with other equipment this scenario / test is valid and should work.

[Torsten.]

Sounds like a wrong cabling of the disk enclosures ... but normally it would be hard to do this wrong. Better check.

[Johan Guldmyr]

See the bottom (p52) of the EVA4400 installation guide to confirm if you've cabled it correctly:

http://bizsupport1.austin.hp.com/bc/docs/support/SupportManual/c01883137/c01883137.pdf

[Uwe Zessin]

Removing both back-end cables is:

1) a double-error. Redundancy deals with single-errors.
and
2) it does not simulate a failure if the controller module.

From what I have seen about the architecture, you cannot really simulate that.

[gregersenj]

Removing all cables from one controller, should not take an EVA down.
Isn't that what you you when you replace a controller?

However, I have seen an older EVA hang during controller replacement.

On the other hand, I have replaced several controllers on EVA's, including the 4400, with absolutely no problem.

If you test again, I expect that it will work.

BR
/jag

[Torsten.]

>> Isn't that what you you when you replace a controller?



No, it isn't.

Because you shut down this controller first.

[linuxbass]

Seems there is not a consensus about how to test controller loss. Have any admins tested for controller failure on the EVA? Should be able to pull cables in my opinion. Why won't it stand up? Let's have some actual technical explanation. That is the aim of the test. Test redundancy in the event of controller failure. How is disconnecting the controller device ports not a proper test?

I would think controllers do NOT do proper shutdowns when the fail.

BTW I used HP's wiring diagram for shelve addition attached here

[Torsten.]

You did NOT test a controller failure, because your controller was still alive and healthy.

[Johan Guldmyr]

There is still the midplane that connects the controllers.

[linuxbass]

But it did lose connectivity to all shelves while having mid-plane connectivity.

why:

1) did the EVA fall over and quit service storage?
2) The CV freeze and stop operating?

What testing is possible for controller failure?

We need to use the EVA for production in a VMware infrastructure. This array CANNOT fail on us or all production VM's will fail.