I finally convinced upper management that we need a computer replacement plan, and I've been tasked with making an inventory of what we have and determining what needs to be replaced this year, next year, the year after, etc. I had to use some sort of criteria to back up my recommendations, so I decided to try using the Windows Experience Index.
I've determined the CPU and Memory scores for all of our desktops and servers using community data. I also feel fairly successful in assigning a WEI score to each user based on their computing needs. I'm struggling with assigning a WEI score to the various servers that we have: file server, database server, Exchange server, backup server (for doing backups), web server. Suggestions would be appreciated.
I think you are misusing the WEI score. The WEI score is not an average it's simply the lowest score from all components in the system that are benchmarked. If you were comparing new machines (and didn't understand the differences between them) this is a reasonable way to tell great grandmother that machine A is better than machine B. It is not meant to determine what type of PC in general you need for a particular task. You can use it to tell users what type of PC is required for any given application. EG Word might need a WEI rating of 3. That doesn't mean that if you are primarily a word user that you only need a rating of 3 for a PC. Likewise if I run 2 applications that require a 3 it doesn't imply that you need a 6. I think you could use it to define subsystem requirements I think, but even then it's not designed to be quantitative just qualitative. EG 1 raid 5 disk subsystem score can be the same as another raid 5 score but the latter might be raid via the bios and the first a dedicated raid card.
As a server is normally run headless the WEI score is really irrelevant, as it's intended for desktop use only. Assess your server requirements based on real need, rather than some fabricated system for impressing users.
Choose a few key indicators; CPU usage, memory usage, disk I/O percentage and measure the existing servers to see what is hitting limits and could benefit from an upgrade.
Talk with the users to see where they are unhappy with performance; if everyone complains that the file server is slow but they love the web-server then upgrades should focus on the file-server, even if the number show the webserver is under higher load.
If there is anything that gives you concerns for reliability, even if it is performing well, consider replacing or upgrading those devices.
Finally if you are going to upgrade remember it doesn't have to be a 1:1 process; you may want to take the chance to combine or split server functions, or potentially even consider virtualisation if that is a better way to meet your needs.
One other thing to consider is that older servers tend to be significantly less power efficient - the computing power per watt of systems that are a couple of generations old can be more than 10 times lower than current generation systems, or even worse. Simply replacing older kit with newer kit can significantly reduce power bills, replacing your existing discrete servers with a consolidated solution using a good virtual architecture running on modern servers can pay for itself in power consumption alone if you are running enough older servers (20 or so seems to be the cut off line in the environments I've assessed).