Very elightened perspective.
My advice is to first ask if your users are programmers or non-programmers. One steadfast rule for a programmer is that he/she needs an operating system.
Can the programmers work with de-identified data rather than a less secure server with identified data,
Pharma and marketing for instance do not need identified data. Programmers do not develop models on 100 million identifiable data. Very rarely is name, credit card number or SSN needed? Develop on limited data then score on server?
Double blind trials are by definition de-identified?
Finally provide statistics on the new platform
1. Reliability (does your program give the results you want )
a. Does the programmer have admin priv so he can apply hot fixes?
b. Can a programmer upgrade to a later version of SAS as soon as it is available?
2. Availability
a. Is there scheduled down time.
b. IBM once guaranteed airlines no more the 15 minutes of downtime per year.
c. If platform is down have penalties users can use to upgrade their power workstations.
3. Serviciabiliy
a. Can users use workarounds either provided by SAS or in other languages when
SAS fails and cannot be serviced.
4. Performance (provide benchmarks for users)
How does the server compare to an inexpensive power workstation with says 128gb ram, dual 16 thread processors, mutiple RAID 0 SSD arrays, three monitors, programmable mouse, command line, programmable mouse, prefix area and command macros,
5 Functionality
1. A matrix language with command line and/or interface to Python and R
2. Command macros, pipes, interactive environment, R Shiny or Django
3. Internet acess
4. Libname engines