EVA4400 Performance Monitor output

[JorisK]

Hi folks,

We think our backup system overloads our EVA4400 SAN. At least, we think it is overloading one of the controllers in which a failover happens and the entire SAN crashes.

I'm currently watching EVAPerf and look at the value 'Read Requests' on all instances (hosts ports FP1/FP2 on both controllers.
What i see is that sometimes one of our ports jumps to 100% utilization, at least... that's what EVAPerf tells me.

How to i have to read this output? Does 100% say it cannot accept any requests anymore and it will queue all other incoming requests?

Regards,
Joris

[Víctor Cespón]

What comand line parameter do you use with EVAperf?

In host por statistiscs (EVAperf hps)

You get:
- Read Req/s
- Read MB/s
- Read Latency (ms)
- Write Req/s
- Write MB/s
- Write Latency (ms)
- Av. Queue Depth

You should look at the latencies.

[JorisK]

Read Req/s on all Host ports.

We normally have four backup queue's running and they cause serious shit every night. Today i started one and noticed during the progress it runs up to 100% utilization for a long period.

Does this mean this port is for a 100% utilized?

[Víctor Cespón]

JorisK: The problem is that EVAperf does not give you any of these values in %. Where are you seeing the 100% utilization?

The performance of the EVA is limited mainly by the type and amount of disks in the disk group. An EVA 4400 has 4 x 4 Gb/s ports, and it's very unlikely that you can saturate a fiber channel port.

[JorisK]

Well the problem is, our backup software has multipath problems, it only uses one path.

The y-axis in the graph goes from 0 to 100, that's why i tought the output would be placed in percents.

[JorisK]

Another question:
How can i determine the maximum utilization of my EVA setup for e.g. controller processor time, host port utilization etc?

[Armin Kerl]

Read Req/s on all Host ports in %
If you got it in % you do not use evaperf.exe direct, you are using Windows Perfmon I think and this says you nothing.
You have to look under the Graph where Perfmon is showing the Values.
Read Req/s are the Disk IO's.
Calculate around 100 IO's for every 10k Disk Drive, this will be the max. IO for your EVA.
Then compare it to the max Value comming from Perfmon.
But If you are running Backup, IO's might not be the interesting Value, I would look at MB/s.