1

We have an VMware structure, that uses a couple of physical hosts, a blade chassis and two SAN storages (a Fujitsu DX90 and a Fujitsu Fibrecat SX88). The two physical hosts have most of the LUNs on the SX88 and the blade chassis as most of the LUNs on the DX90. However, both groups of servers see both SANs.

I was wondering what's the best method to "benchmark" the access to the SANs, as I feel that from the physical servers, the access to the DX90 should be quicker. Right now I'm copying info on a VM with Windows 2008 R2, from a LUN to another on the DX90, and it's giving around 45 MB per second. Since the physical server has to cross two fiber switches, I want to make sure that they aren't "bottle necking" the connection.

Thanks!

Helder
  • 25
  • 5

2 Answers2

1

ATTO diskbench is quite nice for this, as you'll see clearly wich kind of blocksizes your system is optimized for.

They don't have any official download page for it anymore, but I've used this one and can confirm that it's virus/trojan/whatever-free:

http://downloads.guru3d.com/ATTO-Disk-Benchmark-v2.41-download-2343.html

pauska
  • 19,532
  • 4
  • 55
  • 75
1

Don't know about benchmarking with Windows, since I'm a Linux guy. In Linux I tend to use plain old dd command, bonnie++ and dbench.

But here's a note specifically about SANs: in addition to watching raw MB/s numbers I watch latency. With cheaper SANs it's possible that one I/O hungry host negatively affects all the other hosts using that SAN, leading to dramatically growing access times. I've seen access times jumping from couple of milliseconds to several thousand milliseconds and that truly kills the performance.

Janne Pikkarainen
  • 31,454
  • 4
  • 56
  • 78