We are looking to test various iSCSI vendors, but have never done any iSCSI. Our environment is FC to HDS Tagma. Does anyone have any recommendations of what methods or commands to use to compare iSCSI to FC? Hardware or software initiator?
Easiest means of comparing apples to apples (tools wise) is to use iometer or sqlio. Create the LUNs and add them to ESX. I would create a vm and give that vm the same size disk created in all 3 types of storage (FC/iSCSI-sw/hw), and then run IOMeter against those disks.
I've always found FC to come out better with strong throughput and low latency, and it's the latency that usually kills iSCSI for me. With iSCSI, you're only as strong as you're networking infrastructure, so I'm sure you can get good results with it. You won't get much better results, performance wise with SW/HW iSCSI, but hw iSCSI gives you boot-from-san, which sw does not. Also, you're taking the scsi commands off the service console, so that's another benefit of hw. With the new 10G infrastructure, you can get stronger performance, than the current FC implementation, but again, you're only as strong as your network.
-KjB
I agree with the IOMeter. We use this internally to test our own iSCSI systems and compare them between each other. Don't forget to setup the iSCSI on a separate subnet, at the bare minimum. A lot of time people test iSCSI on the LAN and wonder why it is so slow. Good luck with your testing. iSCSI is great for the Virtual Server environment.
Ah, but you can get 8G fiber...and it will be faster than 10G iscsi...
That remains to be seen. Theoretically, of course, just based on speeds, 10G would be faster than 8G, and if your network is setup to do so, FC will get a run for the money. I'd love to see benchmarks if anyone has them.
-KjB
From experience, the low latency, lack of overhead, and packet size....I'm willing to put money on FC.
It's too bad that FC's days are numbered(bond 3 10G links for a lot less than 3 8G fc links no doubt....
Hi,
Testing in a VM can give you results that are far from what the storage subsystem is capable of. An ESX host can limit the resource usage of a VM. I would suggest using native hardware running your OS preference to bench the storage subsystems performance using IOMeter.