Server Error messages re: EVA8000

[Gregg Dow]

Hi There;

We have 10 servers attached to an EVA8000. They ALL began receiving errors about a month ago, (which I just noticed), but we HAVEN'T seen any real problems, or any significant messages (events) on the SAN or the FC switches they're running through.

Windows Server 2003 Event viewer is showing multiple messages like:
"The device 'HP StorageWorks EVA 8x00 Storage Array Controller Device' (SCSI\Array&Ven_HP&Prod_HSV210&Rev_5000\6&398693e2&0&000700) disappeared from the system without first being prepared for removal."

and

"The device, \Device\Harddisk3, is not ready for access yet."

Anyone know what I should be looking at or for to narrow it down?

Thanks.

Gregg

[Jan Soska]

Hello,
could be problem related to EVA or to your SAN infrastructure. Check logs on EVA, if they are ok, focus on SAN switches - check logs, probably upgrade firmware (check EVA firmware too).
Another option: check your MPIO version installed on client servers, FC drivers and firmwares.

Jan

[Rajs]

We have also noticed similar problem with EVA 8000. This we have noticed after a database corruption on a Windows 2003 server.

It is better to verify the EVA and HBA firmware, HBA driver versions, MPIO versions and upgrade them to current and compatible versions.

[Gregg Dow]

Hi All;

Well, it turns out it was a bad fiber run between our two SANs (separate buildings).

Although the servers weren't "presented" to the D/R SAN, they were still indicating problems via Event Viewer. Once the Fiber was fixed the errors stopped. Thanks for all of the suggestions.

Now,.....another question. Do increasing the quantity of LUNS (Virtual Disks)have a performance factor? For instance, breaking up four 1TB virtual disks into eight or more virtual disks,..... does that increase/decrease overhead or not have any affect on the SAN?

Thanks,

Gregg

[Jan Soska]

Hello,

performance usually depends mainly on number of disks - the higher number of disks, the higher IOPS ...

Jan

[Gregg Dow]

Hi Jan;

So the virtualization aspect doesn't matter so much as long as there are enough disks in the DG? We have nineteen 300GB disks in the DG, and 4 large existing luns that we want to carve up into smaller ones.

Thanks.

Gregg

[Jan Soska]

Hmm, hard question.
In our setup we use 2 disk group (12x300GB 10K + 8x500GB Fata), 1 virtual disk per qroup, first vd presented via controller A, second presented via controller B, multipath policy set to RoundRobin with IO count=1 .
We run ~30 VM's including some DB's, performance is very good.

Jan

[Gregg Dow]

Jan;

ok thanks for all of that - I'll wait awhile to see if anyone else answers.

Gregg

[Uwe Zessin]

> Although the servers weren't "presented" to the D/R SAN,
> they were still indicating problems via Event Viewer

They did not have access to any virtual disks, but they did have access to the EVA:

:"The device 'HP StorageWorks EVA 8x00 Storage Array Controller Device'
:... disappeared

This is not a disk - this is a SCSI device type 12(10) LUN, as the texts says: a 'controller device'.

The EVA always presents one at LUN address 0. It is used for in-band management with the server running Command View EVA and often used to ask which disks are presented by the EVA (a more efficent way of polling 256 LUN addresses).


> Do increasing the quantity of LUNS (Virtual Disks)have a performance factor?

Maybe yes, maybe not. It is hard to say in advance without much knowledge about the configuration.


An EVA can deal with up to 2048 outstanding I/Os per controller port. Many SCSI queues on the servers are configured for 8, 16, 32 or 64 I/Os.

If you have more virtual disks/datastores so you have more SCSI queues. And more fine-grained SCSI reservations. If one datastore blows up, less VMs are affected -- to be fair, VMFS is _very_ robust, but errors (e.g. presenting a vdisk to a non-ESX host can happen).

Does it make sense to create one datastore per VM? No, I don't think so. Remember that each datastore has its own free space. If you create lots of small datastores you risk the need to 'shuffle' a lot.

[Gregg Dow]

Hi Uwe;

Thanks, but you're confusing me a bit with VM's/Datastores/ESX. We don't use VMWare, just MS-Clustering at this point.

I'm getting the impression from this thread that if we were to take the 4 large VD's, and carve them up into say 10-12 VD's, that the overhead wouldn't be significant enough NOT to do it.

Thanks.
Gregg