EVA5000 performance bottleneck.

[Sezgin Rafed]

Hi everyone,

We have an EVA5000(2 HSV110CL Controllers) which is being used by around 10 servers.
As far as I know HBA and FibreChannel switches operate at 2 gbps.
My question is:
Where is the performance bottleneck and what is the maximum realistic bandwith ?

To be more specific: if all of the 10 servers are saving a large file to the EVA at the same time, what is the realistic data throughput each of them could achieve ?

Regards.

[Uwe Zessin · Accepted solution]

It is not possible to predict a specific bottleneck - it depends on your configuration if there really is a bottleneck.

The bandwidth is, well, 2 GigaBit per second per front-end port.

Actual throughput depends on your configuration:
- define 'large file' (each controller has 256 MegaBytes of write-back cache)
- how many disks are at the back-end
- how are the disk groups organized
- what is the VRAID level of your virtual disks?
- are you saving through one or all four frond-end ports?

Best for you is to try it out and measure on _your_ configuration. Somebody else can get quite different values.

[dentistcorn]

What can we use to measure the stats on the EVA's controller? Ports, VDISK perf, host connections, etc.

[Anders Trolle]

There is not much performance statistics to get from the EVA itself. But your fibrechannel switches can provide you with some sort of pointer.

If you log on to one of your coreswitches (or the switches to wich your EVA's are connected) using telnet and issue the command "portperfshow", you will see a list of traffic pr. second pr. port.

Our own EVA5000 can produce more than 80 MB/sec without any noticable delay on the connected servers.

[BR824100]

As far as I can see for windows servers there are some eva performance counters as listed in this doc:

http://h71028.www7.hp.com/ERC/downloads/5982-9586EN.pdf

However, I haven't seen these for download anywhere and wouldn't mind a copy of them too!