VMware vSphere vStorage API and HP EVA

[Patrick Terlisten]

Hello,

Round-Robin?! It was always included in VI3, but marked as "experimental". In this case RR is maybe an option.

Best regards,
Patrick

[Uwe Zessin]

What do you mean by "maybe"?
Do you want RR across non-optimized paths, too?

I have not mentioned above, but it is clear that you now should distribute your virtual disks across the controllers from CV-EVA and all ESX4 servers will honor it - no more unexpected ownership transfers due to a wrong settings on one server.

[DDas]

PowerPath/VE supports the EVA 8000 series as well as the active/active EVA 3000/5000.

[Davy Priem]

Is there now any plugin coming for the EVA from HP? When will HP publish best practices for Vsphere, as they did for esx 3.5

[mujzeptu]

Bumped and agreed.

[Ragnar Joensen]

HI

I am also waiting for the HP Best Practive for vSphere.

In the mean time, we have enabled "Round Robin" on for our EVA4000. As mentioned before vSphere is aware which paths are the optimal paths, and only does round robin on those paths. This is rather clever, as I now manage paths from my EVA. I just set the Preferred Controller, and if you monitor your vSphere, you can see that it changes the active paths based on which controller I set as preferred.

[Doug Baer]

For those interested, I believe the 'best practices' wait is over:

http://h20195.www2.hp.com/v2/GetPDF.aspx/4AA1-2185ENW.pdf

With regard to the PowerPath for vSphere option from HP, I don't think this is necessary for the EVA - ALUA is about all that is needed for those arrays, along with the Round Robin option from vSphere.

The XP is a different story altogether.

[Chris House]

Is there any kind of Best Practices for vSphere/ESX 4 and XP family of arrays? I am familiar with the configuration guide but it doesn't specify pathing requirements. I know VMware's compatibility matrix says Fixed pathing is the recommended setting for XP arrays but wanted to see it from HP. Thanks for the EVA best practices guide, we are moving most of our VMs to EVA but still have a sizable population of virtual desktops that will stay on XP.

[John Tokash]

My shop runs EVA's exclusively for virtualization and while ALUA, along with the now available round-robin balances load for performance, it does not however take into account health of the array. What we found in a recent performance analysis of our storage environment was below...

LUN's were set without preferred path - HP recommended we change to preferred path to avoid the EVA trying to decide which controller was best

LUN's were set to FIXED on the ESX host, but without manually assigning a path to a specific controller.

we present all 8 paths to an eva, to the ESX host to present all possible paths for redundancy. Because the ESX host can't talk to the EVA (This is what i'm hoping a v4 API would do) - it can't tell which fabric has the owning controller. an unfortunate downfall of A/A arrays. While in a nominal environment, passing ESX VM IO traffic over the mirror port serves the purpose, once the mirror port reaches near peak conditions - the ESX host suffers increased latency.

I developed a vbscript tool to take the output of a esxcfg-mpath -l cmd from an esx 3.5 host, parse it up and determine which path was appropriate. this based on assingning lun's prefrred path to a controller (odd's to controller a, even's to controller b). this has worked out very well for me, but esx4 takes away the same output from that command.

I am very keenly watching, and asking my HP rep's about an API - as i upgrade to v4 on my hosts, i will revive my performance challenges with the mirror port.

note - my EVA's are not all dedicated to virtual systems, and my high mirror port traffic is driven primarily by physical's, which i'm having a hard time getting adjusted to avoid the mirror port path.

feel free to reach out to me directly/privately if you wish to understand this better. its confusing for me to explain to my storage architects - and this is only a brief explanation.