dan@rna.UUCP (Dan Ts'o) (10/12/84)
Hi, Here is the promised second benchmark article in which I present benchmark results for several more UNIX systems. As in the first article, I have computed normalization indices with respect to a 4.2BSD VAX 780. Thus the benchmarked UNIX system appears as a fraction (or multiple) of a 780. If the program is pure compute, the user time is used, otherwise the real time is used in determining the normalization index. The actual times are given in the end for those who want to know and compare. I also include the normalization indices from the previous article just for reference. If anyone would like to run this benchmark suite on their UNIX system and would like a copy of the scripts and programs used, I would be happy to mail it to them. - LOOP, for loop of 1million with long int index, Same as some previous UNIX conference benchmarks - CC LOOP, cc -O loop.c, Companion C compile to above - SIEVE, Same as published in BYTE - CC SIEVE, cc -O sieve.c - FLOAT, Same as published in BYTE, testing floating point performance *, / - GETPID, for loop of 100000 getpid()'s - GREP, grep zoom /usr/dict/words, grep through ~200kbytes - COPY, cp /usr/dict/words /tmp/junk, copying ~200kbytes - NROFF, nroff -ms /dev/null, load the MS macro package - SORT, sort -r /usr/dict/words > /tmp/junk 780 MVAXI 11/73 Ridge Callan G8750 11/70 LOOP 1 .21 .24 1.3 .40 4.2 .35 CC LOOP 1 .31 .31 .19 .13 1.1 .28 SIEVE 1 .28 .67 1.7 .59 8.0 .86 CC SIEVE 1 .31 .34 .24 .15 1.5 .31 FLOAT 1 .10 .16 .52 .031 3.3 .62 GETPID 1 .32 .35 .21 .89 3.1 .20 GREP 1 .32 .59 1.1 .51 4.1 .24 COPY 1 .25 .14 .57 .15 2.0 .22 NROFF 1 .34 .29 .53 .29 1.4 .46 SORT 1 .33 .44 .72 .47 3.4 .73 Summary of normalizations: mean 1 .28 .35 .71 .36 3.2 .43 standard deviation .074 .17 .50 .26 2.0 .23 Machine configurations: 780: 11/780, Eagle disk on SC780, FPA, 4.2BSD, 4k/1k fs MVAXI: MicroVAX I, Ultrix, (I think RD51 disk) 11/73: USDC 40ms disk with read cache, FPU, no FPA, PWB/Unix Ridge: Callan: Unistar 300, 68010 G8750: Gould Concept 32/8750, 4.1c BSD 11/70: FPA, BRL UNIX (V6/V7/2.9BSD), CDC 9766 Some assorted results: Bench Norm Machine NROFF .41 Fortune 32:16m 30Mb disk NROFF .33 Plexus P35 NROFF .26 Onyx C5012 NROFFMM .34 Motorola Series 2000/260 (nroff -mm /dev/null, see below) FLOAT .41 Masscomp 500 with FPA (see below for result without FPA) COPY .18 3B2/300 GREP .29 3B2/300 Results from previous article: PYR 780 750 11/44 11/34 11/23 MASS 3210 PC/XT 286 LOOP 2.1 1 .49 .27 .19 .1 .38 .23 .080 .16 CC LOOP .6 1 .6 .3 .25 .17 .38 .17 .073 .17 SIEVE 2.5 1 .61 .71 .46 .26 .57 .36 .21 .56 CC SIEVE .67 1 .57 .36 .27 .19 .4 .17 .075 .19 FLOAT .27 1 .76 .31 .27 .034 .030 .33 .13 .0029 GETPID 2.0 1 .59 .41 .30 .15 .76 .25 .22 .55 GREP 1.3 1 .5 .44 .4 .24 .4 .2 .13 .39 COPY 2 1 1 .16 .13 .13 .25 .1 .047 .10 NROFF 1.3 1 .57 .33 .22 .14 .4 no -ms .12 .27 SORT 1.4 1 .55 .42 .34 .20 .5 .22 .16 .41 Summary of normalizations: mean 1.4 1 .62 .37 .28 .16 .41 .23 .12 .28 standard deviation .74 .15 .14 .098 .067 .19 .08 .059 .19 Machine configurations: PYR: Pyramid, Eagle disk, no FPA, running OSx (4.2BSD) 780: 11/780, Eagle disk on SC780, FPA, 4.2BSD, 4k/1k fs 750: 11/750, Eagle disk on SC750, FPA, 4.2BSD, 4k/1k fs 11/44: CDC 9762 disk, FPU, cache, PWB/Unix (512byte/block), 50 kernel buffers 11/34: CDC 9762 disk, FPU, cache, PWB/Unix (512byte/block), 10 kernel buffers 11/23: USDC 40ms disk with read cache, FPU, no FPA, PWB/Unix, 15 kernel buffers MASS: Masscomp 500, no FPA, 4kb cache, virtual memory System III, 68010 10Mhz 3210: 32016 8Mhz, PDP11/23 IOP, 16081 FPU, GENIX (4.1BSD), no wait state mem PC/XT 8088 w/ 8087 FPU, Venix 286 Intel 286/380, 80286 at 6Mhz, no 80287 FPU, XENIX, Priam 3450 35Mb disk Notes and Comments: - All machines were running multiuser with one user. Results presented were reproduced with several trials. /usr/dict/words was confirmed to be of the same 200kb size +- 2kb (1%). The MS macros were not compacted/compiled. - The Gould sure provides multiples of 780 UNIX power, especially in CPU bound programs. - The MicroVAX I doesn't seem so bad in system performance and its right where DEC said it would be. The RD51 disk is slow (85ms access) - a faster disk should help the MicroVAX's system performance. - For a $1300 CPU board, the 11/73 put on a respectable showing. DEC promises a FPA for the 11/73 which should improve floating point performance on the 11/73 by several times. Too bad about its address space limitations. As usual, when DEC says that the 11/73 has "11/44-class performance", that means its slightly less than an 11/44. - I'm not sure that the 11/70 results are quite representative of a properly configured 11/70, though they are probably close. Overall, the 11/70 results are not much better than the 11/44 (or 11/73). - The Callan and the Masscomp seem to confirm that 68000 UNIX's have an efficient system call mechanism. They also confirm the slowness of software floating point. Masscomp has a FPP now which appears quite reasonable in performance. - I do not believe in the system times reported by the Ridge. Luckly, system times do not affect the benchmark. The Ridge seems to have a very slow C compiler and poor system call performance, but quite good integer performance. - I benchmarked the same 11/44 running 2.9BSD and found its performance to be almost identical to the above results using PWB/Unix. Its C compiler performance was 20-30% better, but other specs were slightly worse. These slight differences are likely to be due to differential performance between the I/O systems of 2.9BSD and PWB/Unix. - Just so you folks with System V don't feel left out... The 780 and 750 running 4.2BSD did NROFF using instead the System V -mm macros as follows: 780 NROFFMM r 14.3 u 10.5 s 2.4 750 NROFFMM r 22.5 u 17.6 s 3.6 n .66 while a 780 running System 3.0 did: GREP r 6.0 u 4.2 s 1.3 COPY r 5.7 u 0.0 s 1.9 NROFFMM r 14.7 u 10.4 s 1.7 Thanks to Tim Tucker (ccvaxa!tucker) who provided the Gould 8750 and the MicroVAX I times, Terry Sejnowski (green!terry) for the Ridge times, and Ken Winston (wealth!ken) for the Callan 300 results. Cheers, Dan Ts'o Dept. Neurobiology Rockefeller Univ. 1230 York Ave. NY, NY 10021 212-570-7671 ...cmcl2!rna!dan Appendix of times: 780 MVAXI 11/73 Ridge Callan G8750 11/70 LOOP r 2.6 12 10 2.4 6 0! 8 u 2.5 12.0 10.4 1.9 6.3 .6 7.1 s 0 .2 0 0 .2 0 .1 n 1 .21 .24 1.3 .40 4.2 .28 CC LOOP r 3.4 11 11 18.0 27 3 12 u .7 2.3 .7 5.5 2.7 .2 0 s 1.6 5.0 3.5 .7 3.0 .8 .7 n 1 .31 .31 .19 .13 1.1 .28 SIEVE r 2.6 8 4 1.9 4 0! 3 u 2.4 8.5 3.6 1.4 4.1 .3 2.8 s 0 .2 .1 0 .2 0 .2 n 1 .28 .67 1.7 .59 8.0 .86 CC SIEVE r 4.4 14 13 18.4 29 3 14 u 1.7 5.0 1.8 6.8 4.2 .5 0 s 1.6 5.3 3.8 .7 2.6 .9 .6 n 1 .31 .34 .24 .15 1.5 .31 FLOAT r 1.4 13 8 2.9 41 0! 2 u 1.3 13.1 8.3 2.5 41.4 .4 2.1 s 0 .2 0 0 .3 0 .1 n 1 .10 .16 .52 .031 3.3 .62 GETPID r 18.6 58 53 88.3 20 6 92 u 2.5 7.3 6.6 37.1 1.3 .8 2.5 s 15.8 51.4 45.9 13.5 19.5 6.1 89.2 n 1 .32 .35 .21 .89 3.1 .20 GREP r 4.1 13 7 3.8 8 1 16 u 2.9 11.2 3.1 2.7 6.3 .9 12.3 s .5 1.6 2.6 0 .9 .2 3.3 n 1 .32 .59 1.1 .51 4.1 .26 COPY r 2.0 8 14 3.5 13 1 9 u 0 0 0 .2 0 0 0 s .7 2.4 5.7 .1 1.7 .4 6.2 n 1 .25 .14 .57 .15 2.0 .22 NROFF r 4.1 12 14 7.8 14 3 9 u 2.9 9.4 7.1 3.9 6.4 1.2 4.9 s .6 1.7 3.0 .1 1.7 .6 3.5 n 1 .34 .29 .53 .29 1.4 .46 SORT r 37.8 115 86 52.6 80 11 52 u 33.9 103.7 54.5 26.5 53.8 9.5 36.2 s 2.1 7.3 19.0 1.3 5.9 .9 13.7 n 1 .33 .44 .72 .47 3.4 .73
dan@rna.UUCP (Dan Ts'o) (10/22/84)
Hi, Due to numerous requests, I have posted a copy of the UNIX systems benchmarks used in my recent article, to net.sources. Please let me know of any interesting results. Cheers, Dan Ts'o Dept. Neurobiology Rockefeller Univ. 1230 York Ave. NY, NY 10021 212-570-7671 ...cmcl2!rna!dan