Can RIS be used for a mass deployment?

[Cabales]

Ok, I've been steadily moving towards implementing RIS enterprise wide, but recently we have started to talk about changing desktop images again. This has brought up the question of whether or not RIS could be used to mass deploy the image to about 650 client PCs over a weekend. This is across several offices with a RIS server in each location. The worst of the load would probably be about 150 PCs to a single RIS box. My initial thoughts on this are that is a bad idea since, to the best of my knowledge, RIS does not support a multicast type image solution. However, I thought I would ask here to see if anyone has ever used RIS to accomplish something similar. This would be afterhours with nothing else competing for bandwidth with 100mb bandwidth to the desktops and 1GB bandwidth from the server to the switch. Do you think this is doable or not?

If the answer is no is this something that the new flavor of RIS (WDS?) should be able to handle?

[mmarable]

RIS as well as WDS are initiated from the client. So, if you have the staff to run computer to computer getting the process started, then RIS can do the job. If you were looking for something that you could initiate from a central location, then something like GhostCast would be your best bet.

[Cabales]

I'm more concerned about the bandwdith utilization since it does not support a multicast solution. I don't mind having staff initiate the installation, since the alternative would be staff doing post-ghost tasks, but I'm operating on a relatively tight time frame and I'm not sure if RIS can actually handle deploying all the clients within the alloted time. I know Microsoft claims RIS can handle 75 clients per server at one time, but how realistic is this?

[mmarable]

From my experience with RIS the server and network really only take the hit while in text mode setup. Once GUI mode starts the install is just about self contained in the client. Our RIS box was far underpowered, so we would stage our builds that way. Once a set of machines finished text mode, we'd kick off the next set.

I'd benchmark your server and see how many clients you can do at the same time and how long a given machine takes to complete. You can use that to work out a plan to keep rolling set of "X" number of machines in text mode at any given time. This will give you a constant flow of machines completing and you'll know if your setup will be able to handle the task.

So let's say your setup can handle 20 simultanious builds. Work it out so that as your staff start client #20, client #1 is just wrapping up text mode and has freed up its "slot" for client #21. This way you don't have everyone sitting around waiting for all 20 to finish before they start the next 20.