RSS 

How To: Design a Distributed Provisioning Service environment

Paul asks how he can design a Provisioning Services environment across 1600 sites that must be available in the event of a WAN failure.

Tags: provisioning services ask the architect xd4
Views: 1,628
Rating: 4

Transcript : So welcome to another “Ask the Architect.” And today’s question comes from Paul, and he’s got a challenge: 1600 remote VPN locations worldwide connected to a main data center via cable modem or DSL. Each site 100MB switched to 10 desktops and the servers. Each site has its own DC/file/print server. 4 app servers, they’re all physical, looking to virtualize those. Basically, he has about 20,000 desktops. Only two desktop images used within the entire organization. The changes have been 3 to 4 times per week, pretty small in nature. Site must be…each site must be functional if the WAN goes down. Must support USB video. Can either be physical or thin client devices. Must be able to seamlessly deploy and centralize all the changes to all the sites. That is a challenge. The locations are small in nature. You’re looking at 10 desktops per site. And, truthfully, I’m…this one’s kind of difficult. If we have a data center, and we have 1600 different sites being connected to it, there’s two ways you can go right now with what’s available in XenDesktop. You could do the centralized model, which I’m actually very fond of, because it solves a lot of the distributed computing challenges trying to maintain these environments, trying to push out the…any type of image updates to all the different sites. I mean, it’s 1600 sites. That’s a lot to…that’s a lot of data to be sending across a few times a month. If you centralize this thing, you can have everyone coming across the DSL and broadband links over ICA with the HDX technology. So it’s going to give them…ICA is very thin on the wire. It’s going to give them a very good user experience. Now, the one goal we don’t achieve with this type of a model is that, if you lose that link, that site’s down. And that misses one of your goals. So with a central…if you’re going to do centralized, about the only thing you’re going to be able to do is, you’re going to need redundant links to two different providers, I would recommend. Now, if you would go to a decentralized route, you’re still going to have your data center and all the different sites connected, but you are…you’re going to start having components out here now. You’re going to have a PVS server. It’s going to stream to your local images. Now, one server can easily do hundreds of machines, hundreds of streams. You only have 10 at a site, so this server’s going to be very underutilized. So you virtualize this thing. You say you’ve got 4 servers in there that you want to virtualize. This is one virtual machine. But you’re going to have this PVS server in every location. That’s 1600 PVS servers. That’s a lot. Because you’ve got to maintain these servers. You’ve got to patch these servers. We can update the images. So we would update the image here, in the data center, and then we can then copy down the deltas to the sites. So, if you have a 30 gig image, and you update it with a HA fix, that 30 gig image, probably you’re only going to be looking at transferring 100 or 200 MB of data every time you do an update. And you can schedule these things. So you could do something like WorkFlow Studio to automate this for you so you don’t have to do 1600 different sites manually. You’d have to create some scripts that would, at a certain time, it would create the delta image, copy it down, and then integrate it with the PVS servers at the local sites. So you could automate this. Now, for this use case I’m showing you here, this is using a streamed local desktop. Now, if you wanted to be having hosted local desktops in the site, which I actually…I wouldn’t recommend it, because that’s more components you have to have at the site, and I’m not a big fan of doing that. Then you would require more servers down there, you know, a XenDesktop Controller and a Web Interface Server in the site as well. But it’s two options. You can put PVS server in these locations, virtualize it. You’re not going to…you’re only going to stream 10 devices per site, and then you automate with WorkFlow Studio the delta changes that recur. Or you can go down the centralized model, where you can do a hosted virtual desktop model. Everyone in your offices here would be using thin clients, but you need to have a redundant link, because if the link goes down, you’re out of luck. Whereas in the decentralized model, if the link goes down, they’re still operational. But the problem…the challenge here is, these are not thin clients. These are desktops now. These are full-fledged PCs that you’re going to be streaming the image to. Two different ways you can go. It’s going to be based on what you want to, you know, what you’re trying to achieve, what’s possible within the organization. I hope that answers your question, and, as always, if you ha…feel free to follow me on Twitter @djfeller. And if you have more questions, please send them in to AskTheArchitect@citrix.com. Thank you.