Setting up a VDI infrastructure obviously placed a burden on the network to provide fast response times between the server and client device.
We found in testing that there's no magic formula for figuring out how many client sessions a specific server can support. The number of VDI instances that can be handled by a single server is a function of the server's memory, disk resources, the number of active cores.
Also, not all end users call up applications at the same time, which tends to level CPU load. A number of users all doing scientific math while watching multiple videos, however, can load down each instance.
In the case of a hypervisor-based VDI platform, a hypervisor and/or the VDI management application can have a limit on the amount of impact that each instance may have on overall CPU, disk, memory, and even I/O use.
Depending on use, these resources may or may not be 'oversubscribed/overallocated' because of the aforementioned aperiodic needs among the VMs supporting VDI users. When applications are usable over the RDP protocol, bandwidth needs can be small. The impact of many users running multimedia or graphics intensive programs can increase bandwidth utilization as multimedia sources multiply. There's a fetch to the VM, and a subsequent retransmission of multimedia to the external client 'viewer' that seemingly doubles bandwidth needs at a VM server's Ethernet ports.
Constraining multimedia use for entertainment purposes decreases the need for more bandwidth, but also constrains non-entertainment multimedia applications use.
We found several methods of downsizing Windows XP images by deleting both unused files and killing/removing unneeded processes. The footprint can be further lightened by use of instance slimming software that removes unneeded junk from XP.
NLiteOS and LitePC are examples of products that strip XP instances down to much smaller sizes. Use of pooled disk resources, usually anchored on features of hypervisor platforms, allows Windows instances to share storage in a way that reduces the overall sustained storage footprint.
And as we found in this and prior testing, protocols such as iSCSI can reduce the cost of dedicated SAN fabric by using 'back-channel' storage constructions, if the VDI VM host supports it.