Hello ... can anyone from Dell answer this ?

the MD3000i cannot support a response time limit of 8 seconds at failover. A failover typically takes more than 30 seconds to complete and, while IOs will not be failed, any application that expects a response within 8 seconds will not work.

30 seconds? Maybe... It seems fatser in all honesty. We have critical SQL applications that should bomb if it really took 30 seconds. The controller thing seems real fast to me if not imeadiate, it's the ESX that takes some time and I have found ESX 3.5 to be about as graceful as possible with it.

Hi,
Do you where the settings are in VMware regarding failover ?
thanks,

John

There is some failover stuff in Configuration, Advanced, Disk if your on the esx console GUI itself.

Hi,
Thanks for that, I checked the settings and could only find a timeout for "The number of seconds between FC path evaluations" which I take to mean for Fibre Channel. I will investigate whether or not this applies to iSCSI connections as well.

I ac tually timed a failover on our test setup... ESX / the SQL Server VM was defiently pumping data again at less than 10 seconds. This is esx 3.5 though. not installing 40 till next week.

Ahh .... to be able to afford a test setup ... would be nice. When you go to 4, I would be interested in knowing how quick the failover was and whether or not you had to make any changes to the ESX setup.

cheers,

john

I can report back. I think the 30 seconds is a high side number of a range. I am not sure what factors would slow it down. Maybe the nature of the failure that created the lost path? It may just be data movement, it does seem it would know of a lost path a lot faster if data was moving, rather than idol. When I pulled the iSCSI cable to create the failure I did not have a client running, I was doing a database backup to ensure it was busy at the time of the path loss. Plus from experience I know our SQL servers really hate losing the storage when doing a DB backup.

Did you pull the iSCSI cable from the ESX server ? or from the MD3000i ? I have tried pulling the cable from the ESX server and it uses the other NIC without any loss of service. I also tried it from the switch also without any loss of service. (this was on our ESXi 4.0 development server when all the dev team were at lunch !)
I can't try from the MD3000i as our production servers use it as well.