We're running a WDS server in 2012 (non-R2) in a VM. The network is a mix of Cisco switches, L3 backbones (3560s, I think) and dumb line switches.
When multicasting images from a single instance store WDS is extremely slow, around 35-40 minutes for a basic Windows 7 image to deploy. Multicasting images where there is no RWM takes around 5 minutes.
Unicasting either type of image runs at reasonable speeds for however many systems are imaging. The issue with unicasting is that we'll often image in excess of 100 systems, which tends to make unicasting very, very slow.
I've played with various network settings as found in technet articles or forum posts (e.g. tpCacheSize and so on), and while we've made increases in unicast or non-RWM multicast performance I've not been able to significantly increase the multicasting of RWM files. Both multicast "types" generate similar, and the network guys claim insignificant, numbers of errors. Off the top of my head, when viewing the performance graph for NACKs/Repair/Slowdown Requests, a few hundred NACKs and slowdowns per minute, and 1-2k Repairs per minute.
The network guys configured IPHelpers early in the process, which helped somewhat.
I've found a couple articles on optimizing for HP ProCurve switches, and quite a few on optimizing Windows 2008 (R2) deployment, but they haven't helped me.
Is there a good article on the configuration necessary for Cisco switches to which I could point the network guys? Or has anyone else run into and resolved this peculiar issue?