EVAperf and host port activity.

[Heather O'Neil]

Hi all,

We have started using EVAPerf to monitor our EVA's performance. I have ran into something which is causing concern. I am running the following command our two EVA4400s:

evaperf hps -cont 2 -dur 60 -sz ARRAYName

I constantly see the following:

Read
Req/s
---- -----
FP1 0
FP2 0
FP1 200
FP2 0

Read
Req/s
---- -----
FP1 0
FP2 0
FP1 120
FP2 0

Read
Req/s
---- -----
FP1 0
FP2 0
FP1 452
FP2 0

Read
Req/s
---- -----
FP1 0
FP2 0
FP1 2123
FP2 0

I see similar output for both of the EVAs, namely that the second FP1 is constantly active, while all the other host ports show no activity. This doesn't look normal to me. I have just started using EVAPerf, but it appears to me that there should be some sort of activity on the other host ports as well.

Even if a vdisk is presented through one controller, shouldn't the workload be balanced across the two host ports (FP1 & FP2)?

All of our servers have dual HBAs, and we have dual fabrics.

Any advice is appreciated. Could this be a multi-pathing or zoning issue?

Thanks in advance!

-Heather

[ks6542]

Its depending on constraints such as fabric topology, ISL constraints, server workload skew, and virtual disk workload imbalance.

You need to see those metrics, including those from server performance tools, database internal metrics, fabric monitors, and so on....

[Uwe Zessin]

The EVA cannot 'force' load balancing - it is always the host operating system which decides if it uses one or multiple paths to a single virtual disk (except if there are explicit restrictions made by zoning or vdisk presentation).

Is that a VMware ESX environment? The default setup is that all traffic goes through the first discovered EVA port. Even if the vdisk is managed by the other controller, because ESX3 does not understands what ownership is.

[Heather O'Neil]

Uwe,

Thanks so much for the reply. As of matter of fact, you are correct, both EVAs are currently only serving our ESX 3.5 hosts.

Is there a document or a HOWTO on how to ensure our ESX hosts are load balancing properly? I am no, ESX expert, but I was under the impression load balancing and multiplathing was active by default on ESX 3.5.

I am assuming this is something I will need to take up with our vmware administrators.

Any additional info, would be much appreciated.

Thanks!

-Heather

[Uwe Zessin · Accepted solution]

There is a customer advisory...

HP StorageWorks Enterprise Virtual Arrays - Best Practices for Load-Balancing with VMware and EVA
http://h20000.www2.hp.com/bizsupport/TechSupport/Document.jsp?lang=en&cc=us&objectID=c01796013

It contains a protected link, but if I search for 'c01680588' in a different context I get a reference to:
- HP StorageWorks 4400 Enterprise Virtual Array user guide
514018-001__20090130 (February 2009) [5ed]

(too bad that there isn't even a little Q&A done on advisories, sigh)


Multipathing is always present in ESX, but "load balancing" for a single disk is turned off in 3.5 (last time I checked it was still an 'experimental' feature).

Yes, your ESX administrator(s) need to cooperate with you.

[Magnus Appelgren]

Very late reply...

Uwe, do you know what that document says that are locked?

The open document says:
"no_preference: the load balancing is done from the host side Path A/B failover only. Auto failback is done by the host."

But that options doesn't exist in EVA4400 afaik. Do you know what option to choose in 4400?

BR Magnus