|
Bruce, here is a copy of some mail Doug Tody just sent to IRAF:An interesting benchmark. In both benchmarks, the V2.6 IRAF DISPLAY task was used to display an image
on the (usually) remote node 'pavo', via IRAF networking. The 512sq benchmark
displays a 512 square image of type short (dev$pix) in a 512 square IMTOOL
window. The 1600sq benchmark displays the same 512 square image in a 1600
square IMTOOL window, using bilinear interpolation (fill+) to scale the image
to fit the window. Both benchmarks involve considerable floating point
computation to do a linear mapping and clip, and in the case of 1600sq,
a bilinear interpolation.Machine 512sq 1600sq 512sq 1600sq
(cpu time) (speed relative to the Vax 11/750)orion 1.5 32.0 6.8 5.1 Sun 4, SunOS3.2GAMMA
pegasus 1.9 36.5 5.4 4.4 Sun 3, 25.0 MHz, FPA
draco 3.3 42.2 3.1 3.8 Vax 8600, VMS 4.6
tucana 2.7 44.7 3.8 3.6 Sun 3, 16.7 MHz, FPA
lyra 10.2 161.6 1.0 1.0 Vax 11/750, FPA, 4.3BSD
pavo 7.0 188.3 1.5 0.9 Sun 3/110, 16.7 MHz,
68881 fpu, diskless nodeAll times are cpu seconds. Note that the Sun-4 node (orion) was running the
gamma release of SunOS for the Sun-4, and the timings may well improve in a
future release of the OS. This benchmark tests only the Fortran compiler.
This is only one benchmark, and by itself is not sufficient to draw any
conclusions about the relative speed of these machines. My reason for picking
this benchmark for detailed timings was simply because I had noticed a
significant difference in display load speed on the different machines,
when the fairly expensive fill+ option was used.
|