bruce@tcom.stc.co.uk (Bruce Munro) (09/26/90)
Here at STC we've just acquired some news Sparcstation 1+ workstations
and a SparcServer 4/490. I was demonstrating the relative speed of
these machines against our Sun-3's using a simple benchmark program
based on the Sieve of Eratosthenes.
The program calculates prime numbers using a progressively larger
array. Here's the output from the program for the Sparcstation first,
then the server:
Sieve of Eratosthenes (scaled to 10 Iterations)
Array Size Primes Last Prime BenchTime
(Bytes) Found (Sec)
8191 1899 16381 0.100
10000 2261 19997 0.150
20000 4202 39989 0.300
40000 7836 79999 0.550
80000 14683 160001 1.167
160000 27607 319993 2.400
320000 52073 639997 4.817
640000 98609 1279997 9.800
1280000 187133 2559989 19.767
2560000 356243 5119997 40.367
Relative to 10 Iterations and the 8191 Array Size:
Average BenchTime = 0.121 (sec)
Sieve of Eratosthenes (scaled to 10 Iterations)
Array Size Primes Last Prime BenchTime
(Bytes) Found (Sec)
8191 1899 16381 0.067
10000 2261 19997 0.067
20000 4202 39989 0.167
40000 7836 79999 0.350
80000 14683 160001 0.717
160000 27607 319993 1.767 <---- Large jump
320000 52073 639997 5.150 <---- here.
640000 98609 1279997 11.483
1280000 187133 2559989 24.267
2560000 356243 5119997 49.850
Relative to 10 Iterations and the 8191 Array Size:
Average BenchTime = 0.102 (sec)
On the workstation the benchmark time roughly doubles when the array
size is doubled. This is also also true on the server, with one
exception, when the array size doubles from 160000 to 320000.
Can anyone tell me why this should happen? The behaviour is
consistent, several other runs produced similar results. I'm presuming
that it's some architectural difference, but that's just a guess.
--
Bruce Munro. <bruce@tcom.stc.co.uk> || ...!mcsun!ukc!stc!bruce
STC Telecommunications, Oakleigh Rd South, London N11 1HB.
Phone : +44 81 945 2174 or +44 81 945 4000 x2174
"There are no strangers, only friends we don't recognise" - Hank Wangford