iSCSI datastore over WAN link

[mwpreston]

Just wondering if anyone has any experience with this or if I am crazy to even think about it...

Basically, we have an 8 host cluster that we are currently backing up/replicating with Veeam from an HP EVA 6400 san to an MSA 1500. This is working great, flawlessy. I'm also currently doing some testing of replicating a couple of VM's accross a 100 MBs wan to an esxi box. This also is working great...

I wanted to expand on this remote replication situation and was thinking of puchasing an HP x1500 or something along the same lines, installing it at our remote location, provisioning luns to our production ESX hosts at the central locations via iscsi, and then simply setting up Veeam jobs to replicate some of the mission critical VM's to the iscsi datastore accross the wan?

Does anyone see any issues with this? or will it even work? or will it be just to deadly slow to matter...

Thanks,

[tsightler]

What's the latency on the link? If you're talking about mounting the LUN via iSCSI to a local host across the links, you're performance is likely to me not great unless the latency of your link is very low. Latency is generally more important than bandwidth in this case since when you doing block reads and writes you have to make the entire round trip each time, sometimes multiple times. We have a 1Gb link to a mirror facility about 7 miles away and it has very good latency and mount iSCSI LUN's work pretty well, but for anything with significant distance (and thus latency), no matter the speed, a local server to do the heavy IOP's will likely be faster.

[mwpreston]

Not sure on the latency....I'll see if I can get some numbers...

Currently we are replicating to the same site, however it is being done by a virtualized debian server with some storage attached to it....This seems to work fine....

I guess the only way to find out is to test....What's the best free iscsi solution that you reccomend? I was thinking maybe openfiler in a vm...?

Thanks,

[tsightler]

Openfiler should work great for testing. We opted to simply use Linux targets located at the remote sites. We found this to be quicker and far more bandwidth friendly since the whole "read, write, write" cycle happens locally.