server performances

You do also have to keep in mind the client end, what its specs are, and what type of network it is attached to.

You have roughly the same setup as me, I just have more RAM really. I don't serve AFP from my Xserve I actually have SMB from our primary file storage for imaging Win as well as Mac on a Windows server attached to some SAN.

I wonder what Thomas Larkin does, he has a decent sized deployment, and so does Nathaniel, but perhaps they are more distributed.

Craig E

Our main stress is not the imaging part, I've accepted that it will be slower b/c file copy instead of block copy for the configuration and not being multi-cast. I have three main servers for imaging reasons. The JSS is Intel 8-core, w/8GB RAM, 3 drive Hardware RAID. The NetBoot server is G5 w.8GB RAM and the Master file share is Intel 8-core/8GB RAM, attached to Promise RAID. So we've divieded up the load for imaging to 3 servers, all Gigibit with client on 100MB. (How much is a 48-port gigiabit Cisco switch!?!?!?!?!)

We notice that imaging a couple computers takes about the same amount of time as a bunch. Most we have done is about 20 at a time. The AFP filecopy is limited and the servers never top out.

Not sure if that helps, I'm curious to see if we put the Master fileshare on a Windows box with ExtremeZ-IP (AFP) or SMB if the performance would increase. . .

-Nathaniel
SPPS

We have a 6,200 client deployment. I have 1 JSS which is a dual
quad-core xeon intel Xserve with 8gigs of RAM, and then I have the same
thing with 4gigs of RAM at each building as a master file server and
netboot/image server.

Then I have mac minis set up as fail over distribution points that are
running OS X server. Last year all I had were the Mac minis, and it did
work. Our old JSS was a G5 dual processor with 6 gigs of ram and the
mysql daemon just destroyed that processor. It was always maxed out. Now that I have the new xserve I am sitting between 10 to 20 percent
usage for 6,200 clients. Your biggest bottle neck will always be disk
I/O.

I created one giant image for mass imaging and will go back and do
package based when I get a chance to separate all the packages into
configurations. I just don't have time to do that at the moment. My
image is dual boot and about 9 gigs in size and I can do 20 clients at
once in about 20 to 25 minutes.

I don't think network services will increase performance if you have
slow disk I/O. I do have RAID 0 running on my Casper servers.

Running OS X 10.5.4 server, Casper 6.01

We have a deployment of about 4000 macs and this is what we use:

JSS: dual proc-quad core 2.8GHz w/8GB RAM on 10.5.4
Netboot/File serving: Dual G5 2.0GHz w/8GB RAM on 10.5.4
2 additional Distribution points - one powerMac and one windows blade server running ExtremeZ-IP on a SAN.

Network - gigabit fiber between schools and between closets and switches, 10/100 to the desktop. We have a couple all-gigabit switches that we will bring around for special imaging projects.

We max out the main package server at about 120MB/s when imaging a bunch of Macbooks on a gigabit connection which is probably about the realistic maximum. We never got close to this kind of performance on 10.4 and 2GB of RAM in the netboot/package server. I think the RAM is the key. AFP will use all the RAM you can throw at it. When imaging a bunch of iBooks in a gig switch (they are 100 only), we usually max out around 80MB/s or so.

The JSS just cruises along for us. It's most busy when we're either doing a lot of imaging, or first thing in the morning when everything is turning on. For scheduled tasks, we run a startup and then a 30mins scheduled task to give some idea of load there.

-John