|
|
|
Forum Newbie
Group: Forum Members
Last Login: Thursday, July 31, 2008 9:27 PM
Posts: 1,
Visits: 2
|
|
I'm looking at various storage options for a deployment of virtual machines. Currently what I'm looking at is running about 10 racks with 4 c7000 blade chassis in each rack. Each blade chassis would have 4 SB40c storage blades with 300gb drives(when they come out). So i'd be looking at possibly a cluster of 160 VSA devices. Is something like this even possible? I would probably end up splitting it up to 5 clusters of 32, or basically 1 cluster per every 2 racks, since 1 virtual center can't support more than about 100 hosts.
Have deployments of this size been tested?
|
|
|
|
|
Supreme Being
Group: Moderators
Last Login: Yesterday @ 2:22 PM
Posts: 101,
Visits: 354
|
|
With physical SAN/iQ nodes, yes, with VSA, no tests I can speak about.
VSA is mostly intended for small environments that are just below the realm of using a real SAN. To oversimplify the description I'd say <40 VMs and <6TB of storage.
Every day somebody builds environment bigger than that though and shows that is works well. Just a difference of what we intended it for and where customers are taking it.
I'd classify what you describe as big enough to require the real SAN/iQ NSMs though.
That said I'll tell you what sounds more reasonable if you want to pursue this route (this is the VSA forum after all).
Others have similar ideas but I've not seen it in production.
Clusters of 32-40 are possible and have been tested. I'd recommend a max of 16 though for VSA. I think the largest VSA cluster I've seen in use in production is 8 nodes. Keep in mind you can have as many management groups and clusters as you want and manage them all from the same UI simultaneously. Splitting up into many clusters does not have any downside.
Your biggest bottleneck is going to be bandwidth between nodes across blade chassis. Given that a really good idea might be to create clusters that subsume an entire blade chassis, and each chassis has its own cluster. That way you use the super fast blade internal network and not hit the typical bottle neck where the network leaves the blade as much.
Adam C
Product Manager
LeftHand Networks
|
|
|
|