One or two CPUs the pros & cons 3817

Brian Westerman

you can easily see in the LSPR numbers that as the number of processors increase ... the relative thruput multiplier (as a function of a single processor) declines (i.e. the LSPR multiplier divided by the number of processors).

the trivial corollary then is that the incremental increase as a percentage of a single processor ... is also declining (and has to decline faster than the overall decline ... or otherwise the simple arithmetic showing decline in overall thruput as a function of the number of processors ... wouldn't be happening).

ref:

take some of the LSPR ratios listed in this reference

and do some trivial arithmetic ... divide the overall thruput multiplier by the number of processors and see that the percentage declines as the number of processors increase. in order for that percentage to decline the incremental thruput increase contributed by each additional processor has to be decreasing.

i.e. from previous post

from some trivial examples in

in the PDF version of the reference document ... see table 5-1 LSPR table Z990 on (PDF) page 177 ... or document page 164.

going from one processor to two processors has thruput factor going from 1.0 to 1.9 (of a single processor) ... actually low of 1.8 to a high of 1.95.

if you divide the two-processor 1.9 thruput factor by the number of processors, you get .95.

by the time you are at 32 processors the thruput factor is around 20 (for different workloads it ranges between 14.17times to 24.18times). if you divide the thruput factor 20 by the number of processors ... you get approximately .60. if you calculate the difference in thruput increase between 31 processors and 32 processors ... you find that the amount of thruput increase of adding the one additional processor is equivalent to less than 40percent of a single processor.