Article 1085 of alt.sys.pdp10: Path: shellx.best.com!news1.best.com!sdd.hp.com!spool.mu.edu!agate!library.ucla.edu!newsfeed.internetmci.com!tank.news.pipex.net!pipex!in1.uu.net!news3.digex.net!digex.net!not-for-mail From: doug@ss2.digex.net (Doug Humphrey) Newsgroups: alt.sys.pdp10 Subject: Re: Interesting question... Date: 17 Sep 1995 17:57:27 -0400 Organization: Express Access Online Communications USA: 800-969-9090 Lines: 63 Message-ID: <43i5k7$f3d@ss2.digex.net> References: <811240514snz@rwentk.demon.co.uk> <43elu6$36@caesar.ultra.net> NNTP-Posting-Host: ss2.digex.net In article <43elu6$36@caesar.ultra.net>, >Richard Wentk wrote: >>In terms of speed and power - never mind OS wars, let's just talk >>about bare metal here - how does the KL-10 compare to the PC I have >>under my desk? This thing clunks along at 66MHz, has 8Mb of RAM >>and 1Gig of storage. That would have been a mighty hefty computer >>a decade ago, no? >> >>Any comments, anyone? An issue that has been debated a thousand times - the main difference, and the reason why it is difficult to compare the machines, is the matter of "system architecture" and of course what you were using the machines for. Your PC is N mips. The KL 10 was between 1 and 2 mips as I recall. My old KA was way-sub-mip. Yet, KL10 systems were spotted with hundreds of logged in users, and keeping the users happy. Logic states that it must not be a matter of CPU speed, and this is correct. "System Architecture" is a lost art. "Processor architecture" is the new art. One does not replace the other. We have PCs here at Digex that run BSD and play nameservers, and they are great, but they are loaded with about 100 megs of memory to make sure that they never, ever, ever, ever swap, because then the IO system of the PC gets involved, and it sucks SO BAD that it ruins the utility of the machine. As long as you are confined to the kernal, no little context switching, no IO, these things blaze! As soon as you need to get into the context switching/paging/swapping game, down the drain... DEC10 system architecture was pretty good, and it had to be because you couldn't just assume that the CPU was so fast that you would let it do all of the work. Smart comm front ends (from the DN87 and growing up from there) handled issues of interupts and tty service; smart channels talked to IO devices and took care of moving blocks of data around without the CPU getting involved very much; memory subsystems did their best to give fast, wide, wait-free delivery of data to the CPU, and to other devices like the channels (you didn't have to go through the CPU to get to main memory - some people who only know "modern" machines might not understand that as even being possible). In short, a MIP went a whole lot farther when all of the other pieces of the puzzle had at least a little bit of smarts in them (they had logic, but not CPU's since there were no micro-processors yet) and could be told "go move that block of data from that IO device into that block of main memory, using the dedicated memory buss (and the fact that the core memory boxes could talk to 4 or 8 different ports (i.e. devices) all at once) and then interupt me when you are done; I'm gonna go run this program over here while I am waiting for you..." You know, nobody seems to be building mainframes any longer. Maybe if you decided to apply some of the old systems architecture rules to the current technology... No, then the machines cost real money, and all machines these days are "dumpster upgrades"; you throw them in the trash when they break or you are done with them. I enjoyed the 10 architecture. Doug