Active/Active EVA8000 OpenVMS 7.3-2

[Karen Lee_3]

I have the possibility or accessesing our new EVA8000 - will my OpenVMS 7.3-2 be able to negotiate the Active/Active disks?

[Hoff]

The "Multipath Enhancement for Active-Active Feature of EVA and MSA Controllers" appears to be what you want, and it's a V8.3 feature. Haven't looked to see if it's been back-ported.

OpenVMS Alpha V8.x isn't a major upgrade for an OpenVMS Alpha V7.x release.

[Karen Lee_3]

my eva group presented a disk from the EVA8000 and I can mount it a write to it without a problem.

Is the enhancement just improvement in IO?

[Dave Gudewicz]

I've seen comments about OpenVMS v8.x not being a major release for Alpha. I believe it but other who read the official words on this subject think otherwise.

For many shops, there's a big difference in paperwork/approvals when moving from a v7 something to a v8 something.

For a reference on this subject:

http://h71000.www7.hp.com/openvms/binary_compatibility.html

Dave...

[Uwe Zessin]

> Is the enhancement just improvement in IO?

Correct. The EVA 4000/6000/8000 provide asymmetric active/active support. It means that there is still one controller who owns a single virtual disk at a time and who does the I/O to the disk drives. If the non-owning controller receives an I/O request, it passes it on to the owning one - on the EVA 3000/5000 with active/passive failover it would have responded with a "NOT READY" status.

OpenVMS was able to deal with A/P, but it was not able to find out the "performance path" on an A/A/A array. Of course, you were able to select it manually if you were familiar with the details...

[Ian Miller.]

VMS Alpha V8.2 is a major version update as far as the version number goes. However the internal changes between V7.3-2 and V8.2 are not big so its more like V7.4

[Robert Brooks_1]

I was directly involved with the investigation for EVA Active/Active support within VMS Engineering.

The I/O performance difference is *only* on read I/O (bearing in mind that most VMS systems have a read/write I/O ratio of 75/25, or higher).

The change made to support Active/Active on the various EVA controllers was contained in the SCSI Disk class driver SYS$DKDRIVER for V8.3 and was not backported
to any older version.


-- Rob

[Hoff]

I'm likely the guy that wrote the text that led to that binary_compatibility.html document, FWIW. That text looks really familiar.

The text is correct in as far as it goes, but there are other cases where a version number gets bumped for other reasons. OpenVMS Alpha V8.0, for instance, was not a traditional upgrade. V7.0 was a major release, however.

The V6.x to V7.x upgrade on OpenVMS VAX was not a traditional major release upgrade, either. V7.0 may well have arguably ended up as OpenVMS VAX V6.3.

There's potentially more automated approach around verification. It involves mapping the kernel API versions against the image header references for same. (I should create a VerifyUpgrade tool to do this; that would not seem particularly difficult, but would seem useful.)

There have been kernel changes in mid-release, there was a TQE structural change around V7.3-1, for instance -- the old linked list became a tree structure at V7.3-1. So you can find kernel changes even in minor and even in dash releases.

Personally, I'd expect OpenVMS VAX V7.3-2 to V8.x to be equivalent to a minor upgrade.

[Robert Brooks_1]

Personally, I'd expect OpenVMS VAX V7.3-2 to V8.x to be equivalent to a minor upgrade.

----

Steve, of course, meant OpenVMS Alpha; there is no planned OpenVMS VAX V8.x release.

I'd also agree that V7.3-2 --> V8.x is not as potentially treacherous as the traditional major version upgrade.

-- Rob

[Uwe Zessin]

> The I/O performance difference is *only* on read I/O

Yes, sure. For write I/O the data always goes across one of the mirror ports. If the I/O goes to the non-owning controller, it is sent to the owning one to do the physical I/O. If the I/O goes to the owning one, it is set to the non-owning controller to be saved in the cache-mirror.