Path: utzoo!attcan!uunet!mcsun!inria!mirsa!jlf From: jlf@mirsa.inria.fr (Louis Faraut) Newsgroups: comp.arch Subject: RISC vs CISC simple load benchmark; amazing ! Message-ID: <8019@mirsa.inria.fr> Date: 11 Jun 90 12:06:38 GMT Reply-To: jlf@mirsa.inria.fr (Jean-Louis Faraut) Organization: ESSI, Sophia-Antipolis (Fr) Lines: 117 Hi everybody, here is a posting from France ! *First, excuse me for bad English, I'm *NOT* an English wizard ! It is a copy of a previous posting I sent to comp.unix.wizards. I received many answers from it, from which someone pointed out that comp.arch should be a better newsgroup to post this stuff. So I think I have to put it here to get still more feedback. ****BEGIN****BEGIN****BEGIN****BEGIN****BEGIN****BEGIN****BEGIN We are at present in the process of changing ours school old VAX8650 to a more powerful and less expensive machine :-) Therefore, we are looking at new RISC technology and how big CPU manufacturers announced performances are is a matter of wondering for me :-? Nevertheless, I'm a bit circumspect ... So I decided to write a little benchmark program and I ran it on some RISC machines . In this posting, I'd like to submit you some recent and somewhat surprising results I've got from it :-( Purpose of my bench was to establish a RISC vs CISC comparison under heavy CPU loading . For all that, I wrote a general program called "multiuser" which simulates heavy loading by running many simultaneous Unix commands then getting back times measured by "getrusage" and "gettimeofday" system calls . I try it with different commands to test CPU, I/O etc ... but for the sake of brevity, I'll only present here results obtained from a slightly modified version of the famous Bill Joy's test program, where RISC are supposed to be better than CISC . Here is my version of the Joy's program : ======================================== #!/bin/sh echo 99k65536vvvvvp8op | dc ======================================== and following are the results (all times in seconds, ratios computed by the Spreadsheet Calculator) ------------------------------------------------------------ simultaneous commands # 1 2 4 8 16 32 VAX 8650 (8 mips CISC) Ultrix 1.2 Joy user 0.02 0.02 0.05 0.11 0.21 0.48 system 0.04 0.08 0.19 0.36 0.77 1.58 real 0.25 0.33 0.53 1.12 2.31 4.59 DEC 5810 (18.7 mips RISC/MIPS) Ultrix 3.0 Joy user 0.01 0.01 0.08 0.10 0.11 0.46 system 0.10 0.18 0.27 0.66 1.27 2.33 real 0.14 0.21 0.39 0.83 1.54 3.07 user/VAX 2.00 2.00 0.63 1.10 1.91 1.04 average=1.45 system/VAX 0.40 0.44 0.70 0.55 0.61 0.68 average=0.56 real/VAX 1.79 1.57 1.36 1.35 1.50 1.50 average=1.51 SUN 4/390 (16 mips RISC/SPARC) SUN OS 4.0.3 Joy user 1.74 3.52 7.04 14.00 28.29 57.08 system 0.10 0.22 0.41 0.88 1.68 3.76 real 1.93 3.78 7.51 14.99 30.16 61.37 user/VAX 0.01 0.01 0.01 0.01 0.01 0.01 average=0.01 system/VAX 0.40 0.36 0.46 0.41 0.46 0.42 average=0.42 real/VAX 0.13 0.09 0.07 0.07 0.08 0.07 average=0.09 ------------------------------------------------------------ DEC5810 exhibits a 1.5* VAX power -not so good indeed !- but main surprise comes from SUN SPARCServer 4/390 that features for this test 1/10th the VAX 8650 performance while rated at twice the power ! I'll take care not to conclude that SPARC is bad, because this simple test does not pretend to check all the aspects of CPU performance measurement, but as I strongly rely that there is no gross bug in the source, I think one should be prudent before coming to a decision . ... Also one says : "There's no smoke without fire" . ****END****END****END****END****END****END****END****END****END Please give me some feedback about these results, but before you proceed to answer, let me add some details : -1) tests were run singleuser -2) I tried to port the Ultrix 1.2 "dc" command source code to the SUN but I have not noticed any significant time difference with the SUN 4.0.3 "dc" command execution, thereby conditions of tests appears to be the same. -3) Someone argued that I cannot compare RISC and CISC . Well ! But I *HAVE* to do so, before moving from RISC to CISC, because I can't replace that old big VAX by something that cannot support many students (100+) at the same time ! Interested in getting multiuser program source ? Send me an electronic mail . If there are enough requests for it, I'll post it to the net . *Disclaimer : I'm not DEC remunerated . @ , ,, ,,_._. / // // Jean-Louis Faraut / // //-- // / // // Administrateur Systeme ((_._' ((_._. // de l'ESSI E-mail : +-----------------------------------------------------+ jlf@cerisi.cerisi.fr | ESSI (Ecole Superieure des Sciences Informatiques) | jlf@mirsa.inria.fr | Sophia-Antipolis (France) | Tel. : 93 95 44 37 +-----------------------------------------------------+