I agree with you on that. Capacity planning is an art in itself, and a competent planner might be able to take all things into consideration after spending some time in different stations and scopes (sys admin, dba, sys engineering, security, etc). Having worked with both weblogic and oracle, I'd typically fall back on past experiences to guage hardware sizing, and slowly scale up based on intutitve tools (MRTG to look at memory usage, elapsed time for average transactions, etc).
It becomes more difficult with SAS because *a) its not apparent from typical resource monitoring tools (e.g. perfmon.msc, MRTG, Rational Robot)that an application is underperforming or is constrained by hardware (I've read the sugi paper 277-26, 276-26 and the other paper for server 2003), and there arent many clear benchmark units (e.g. elapsed time per transaction given per unit of data). Even for data management, I've been peering through quite a few white papers on I/O optimisation, and some specifically on wintel (not sas related). I'd like to depend on the vendor but they arent very forthcoming, and I am personally disappointed with the quality of deliverables (granted that they are also understaffed, overworked and covering too many product suites).
At the end of the day, I'd love a vendor or consultant to come in, take a look (maybe ARM, maybe something else), and say "alright, you should have 4 x 3GHz cpu and 4TB of disk space, 16 GB of physical memory, using /pae if you want to accomplish this much work for this many users" but its not going to happen. Hence, I feel as though I am slowly postulating trends and requirements by looking at SAS as a black box, and making assumptions that I cant validate. Thats very dissatisfying for an administrator who wants to ensure his hardware and systems architecture is best suited to the application suites running on it.
Interestingly, I concur that a huge part of this is getting support from other stakeholders. Each IT procurement or deployment is there to meet a business need, but if stakeholders dont see the benefits, they have no reason to pay for training or additional memory.
Message was edited by: Joshua