E.g. you may be amused to know that the total NSF Centre usage during the period Oct-94-Aug 95 (11 months) is quite well approximated by
N(t)= 20 exp(-t/80000) + 100 exp(-t/10000) + 210 exp(-t/1000)
where N(t) is the number of PIs whose use during the period exceeded t service units (1 XMP hour or 1.25 SP2 node hours). This corresponds to about 2.8MSU (=20*80k+100*10k+210*1k), which is about 75% of the total usage during the period. As seen from Fig. E3 in the usage statistics, the academic use corresponds to about 75% of the total usage. The remaining time is distributed over industry using about 5%, research staff at the centres, about 15% and 5% other usage.
From CTC, I received detailed from statistics from on day in december, showing that, although the larges number of jobs was running single-node, these jobs accounted only for a very small fraction, and most resources were used by jobs using 64 or more nodes, as shown in the Figure
Analysis of additional NSF data will be added,
as I work through them.
"I keep six honest serving men. They thought me all I knew
Their names are What and Why and When and How and Where and Who"
(Rudyard Kipling, cited by Raj Jain, in The art of computer system performance analysis)