Theoretical disk throughput

Theorestical or practical?
125MB per connection is Theoretical.

Realistically? Ran a migrate of the storage area in 4.1 and you will see what it's pratical is.

On the Configuration, Storage tab make sure to change each lun to Round Robin or you only get one path. *Properties, Manage Paths"

You may be able to do bonding as well, but generally? Once you get lots of activity going from different hosts to multiple datastores? Round Robin is actually very efficient at providing max bandwidth potential. So bonding only buys you a lot when testing say one VM running.

It varies a lot, I have seen people do some pretty amazing things with that little array in the past. Most folks individual ISCSI connections performan somewhere around the 90MBps range, Jumbo frames are a lot more efficient but you'll still typically have a hard time getting really close to the 125MBps theoretical maximum.

In terms of IOPS you can get a lot out of a single disk in this case I used toshibas specs and put it in an iop forumula 1/ [ 0.0029 + (0.0040 + 0.0044)/2 ] = 161 iops, that's a maximum but you'll never see anywhere near that because of raid and various other reasons. I would count on realistically getting around the 100-110 iop range out of each disk. http://storage.toshiba.eu/cms/en/hdd/enterprise/product_detail.jsp?productid=361

Also caching plays a pretty big role in this, it's all pretty hard to factor in though, the post above me is right though you just need to get on the system and try it out, if the numbers even relatively are close to what I posted above you are probably in the right range.

Hope this helps,
~Daniel

Thanks for the info, I'll relay that back to the powers that be and see if they will be happy with those figures. I guess the only way to give them practical figures if they ask for them is to jump in and see but if I get close to the figures suggested above I may actually have the config about right.

I got the Round Robin path configuration from some heavy reading online but it's nice to have it confirmed as the right choice.

What iops setting do you use on the devices? One of the other guys here reckons it should be set to 1 but I can't find anything to confirm this as the right choice.

"Thanks for the info, I'll relay that back to the powers that be and see if they will be happy with those figures. I guess the only way to give them practical figures if they ask for them is to jump in and see but if I get close to the figures suggested above I may actually have the config about right.

I got the Round Robin path configuration from some heavy reading online but it's nice to have it confirmed as the right choice.

What iops setting do you use on the devices? One of the other guys here reckons it should be set to 1 but I can't find anything to confirm this as the right choice."

Tough to really test it from the VM.... IOMETER is OK... But your guest hardly ever will demand the max available period.

A storage migration where the same block size is used on the two stores involved will really show you what can be pushed. Those numbers are usually darn impressive.