EVA4400 + CV9.1 failing big time

[IvanForceville]

We have two EVA4400's in a CA setup with two CV9.1 servers.

Since the start of the project with had numerous problems with CV-EVA losing connection with the EVA's. Apparently a known problem with this setup.

HP promised to release a new FW for the EVA's where the master/slave controller issue and the mismatch in CV-EVA should have been resolved.

Unfortunatley, the issue still appears and the only way is to reboot the controllers in the EVA to get CV-EVA back on track again.

We're running FW09522000 on the controllers

Until now are experience with the EVA4400 is very bad. Are we the only one facing this issue?

[jpmartin · Accepted solution]

Hi,

i had same problems in a similar configuration and is now resolved with a lot of thinks and especially a very fine granularity in zoning practises, as seen in https://forums13.itrc.hp.com/service/forums/questionanswer.do?admit=109447627+1260082947450+28353475&threadId=1376953

The zoning practises are in the 1st fabric :

for the management CV server:

- CV1_1,EVA1_C1P1,EVA2_C1P1
- CV2_1,EVA1_C1P1,EVA2_C1P1

for CA replication:

- EVA1_C1P1,EVA1_C2P1,EVA2_C1P1,EVA2_C2P1


and for Hosts I/0 :

- HOST1_1,EVA1_C1P1
- HOST1_1,EVA1_C2P1
- HOST1_1,EVA2_C2P1
- HOST1_1,EVA2_C2P1
- HOST1_2,EVA1_C1P1
...

and the same method for 2nd fabric.

Hope this help you

Jean-Pierre

[Michael Leu]

I can feel your pain, just last friday we had to reboot the master controller of one EVA4400 because of this issue with XCS 09522000.

HP support has also told us that it should be fixed in the next XCS release.

At least identifying which EVA was causing the CV lockups has gotten easier since I learned about 'cvutil paths'...

[IvanForceville]

Michael, Do you have any specific zoning configurations like jpmartin?

Our zoning is like this (more or less recommended by HP, and not in the books)

REPLICATION_ZONE for Fabric1
EVA1_CTRL1_FP1
EVA1_CTRL2_FP1
EVA2_CTRL1_FP1
EVA2_CTRL2_FP1

CV-EVA Zone Zone for Fabric1
CVEVA_HBA1
EVA1_CTRL1_FP1
EVA1_CTRL2_FP1
EVA2_CTRL1_FP1
EVA2_CTRL2_FP1

Since I have a case open @ HP I will check what their recommendations are.

JPM, thx for the clarification about the zoning.. keep me posted.

[Michael Leu]

We don't have any CA on our EVA4400.

Our zones always have ony two members, like this:
Zone CV_P1_EVA4400_C1P1 for Fabric 1
-> CV_P1,EVA_C1P1
Zone CV_P1_EVA4400_C2P1 for Fabric 1
-> CV_P1,EVA_C2P1
Zone CV_P2_EVA4400_C1P2 for Fabric 2
-> CV_P2,EVA_C1P2
Zone CV_P2_EVA4400_C2P2 for Fabric 2
-> CV_P2,EVA_C2P2

[Uwe Zessin]

I would replace this:
> - CV1_1,EVA1_C1P1,EVA2_C1P1

with:
- CV1_1,EVA1_C1P1
- CV1_1,EVA2_C1P1

because the former creates a 'hidden' CA communication path. What if you want to stop CA traffic for some reason?


Here is the advisory that requests single-initiator/single-target zones, at least for management:

c01822148 - HP StorageWorks EVA4400 - HP StorageWorks Command View EVA V9.1 SBM Losing Connection to EVA4400 After Upgrade to XCS 09520000
http://h20000.www2.hp.com/bizsupport/TechSupport/Document.jsp?lang=en&cc=us&objectID=c01822148

[jpmartin]

Hi Uwe,

in the HP Support Document c01822148, they relate to spit One Initiator with One Target, instead One initiator with Two targets in the case of addressing one EVA.

It is exactly i have done, but in the case of a CV server adressing two EVAs in a CA configuration, the CV server must have in the same zone the two EVAs.

Trerefore, i have spit One Initiator with Two targets, instead One Initiator with Four targets for the zoning management.

For the Hosts I/O servers i have done the One Initiator with One targets schema.

[Uwe Zessin]

> the CV server must have in the same zone the two EVAs

Must? In the past I have made two different zones, e.g.:

CV1_1,EVA1_C1P1,EVA1_C2P1
CV1_1,EVA2_C1P1,EVA2_C2P1

That worked fine and avoided a hidden CA path.

[jpmartin]

I tried this a time several years ago, but when i had to create a DR group, this did not work.

So, several years ago, we did not create CA replication zones

[Uwe Zessin]

Maybe it is not obvious, but I _do_ create separate zones for EVA-to-EVA replication and it (single-server/single-EVA zones *plus* separate CA zones) _does_ work for me.