Article View: alt.folklore.computers
Article #269259Re: A Dark Day...
From: Ken Moore
Date: Fri, 06 Jun 2003 07:52
Date: Fri, 06 Jun 2003 07:52
38 lines
3156 bytes
3156 bytes
In article <bruce-AA539B....@copper.ipg.tsnz.net>, Bruce Hoult <br...@hoult.org> writes > >[1] manufacturers of microcoded machines frequently made low end models >with limited hardware resources and narrow microcode instructions, and >higher end models with more hardware and wider microcode instructions. That reminds me of my days developing a scheduling system for small jobs in a scientific computing service. The ICL 1907 had the same fixed point hardware as the 1906, but had floating point hardware where the 1906 used an interrupt to kernel code. One morning, I found that my test run (after some attempt at an improvement of my code) actually took about 15% longer in clock time, though the computer-measured CPU time was the same or slightly less. I reran the unaltered version and got the same result. At this point I asked the maintenance engineers what they had changed. At first they denied changing anything, but then realised that I knew they must have changed something, and then said that they had found that the floating point unit gave wrong answers for some arguments to one of its instructions (either FIX or FLOAT), so they had bypassed it and inserted the software emulation from the 1906 to perform it, in the confident assumption that a) this instruction was rarely used, b) no one knew what the clock time for execution of a program was and c) the CPU time spent in the kernel was not reported to the logging system. Unfortunately for them, a) we were running Extended Mercury Autocode, which used this instruction heavily, so I had an EMA compilation and execution in my test set; b) the kernel call took about 500 microseconds; and c) I was interested in reducing disk access, card reading and line printing times, so I used a stop watch in my tests. (Also I didn't trust their logging system). The main thing I learnt from that was the deviousness of manufacturers' staff. -- Ken Moore k...@mooremusic.org.uk Web site: http://www.mooremusic.org.uk/ I reject emails > 300k automatically: warn me beforehand if you want to send one
Technical Details
Message-ID:
<6MhnxLAmgE4+EwGH@mooremusic.org.uk>Path:
mbox-mo1.pugleaf.net!archiver1.google.com!news1.google.com!newsfeed.stanford.edu!cyclone.bc.net!skynet.be!skynet.be!fu-berlin.de!uni-berlin.de!ppp-0-201.manc-b-2.access.uk.tiscali.COM!not-for-mailReferences:
<ba5se1$92e$1@news3.bu.edu> <%X3Da.108857$cO3.7881584@bgtnsc04-news.ops.worldnet.att.net> <bbisck$8re$1@pegasus.csx.cam.ac.uk> <3EDE83F0.E4FF8B70@yahoo.com> <bruce-AA539B.18270905062003@copper.ipg.tsnz.net>Show Full Headers
Complete Article Headers
Path: mbox-mo1.pugleaf.net!archiver1.google.com!news1.google.com!newsfeed.stanford.edu!cyclone.bc.net!skynet.be!skynet.be!fu-berlin.de!uni-berlin.de!ppp-0-201.manc-b-2.access.uk.tiscali.COM!not-for-mail From: Ken Moore <k...@i12.com> Newsgroups: alt.sys.pdp10,alt.folklore.computers,comp.arch Subject: Re: A Dark Day... Date: Fri, 6 Jun 2003 08:52:06 +0100 Organization: Moore Music Lines: 38 Message-ID: <6MhnxLAmgE4+EwGH@mooremusic.org.uk> References: <ba5se1$92e$1@news3.bu.edu> <%X3Da.108857$cO3.7881584@bgtnsc04-news.ops.worldnet.att.net> <bbisck$8re$1@pegasus.csx.cam.ac.uk> <3EDE83F0.E4FF8B70@yahoo.com> <bruce-AA539B.18270905062003@copper.ipg.tsnz.net> Reply-To: Ken Moore <moore...@i12.com> NNTP-Posting-Host: ppp-0-201.manc-b-2.access.uk.tiscali.com (80.40.42.201) Mime-Version: 1.0 X-Trace: fu-berlin.de 1054886656 12204558 80.40.42.201 (16 [144353]) X-Orig-Path: mooremusic.org.uk!ken X-Newsreader: Turnpike Integrated Version 5.01 U <RXeoiskBJF+S9YXukCFlKIdARX> Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit