davidsen@steinmetz.ge.com (Wm. E. Davidsen Jr) (03/25/89)
In article <28819@bu-cs.BU.EDU> bzs@bu-cs.BU.EDU (Barry Shein) writes: | 5. Fujitsu claims they will be producing 64Mbit memory chips in a | couple of years. This means a 16Mbyte workstation, with the same chip | count, becomes a 1GB workstation. Does anything need to be evolved to | utilize this kind of change? Is it really sufficient to treat it as | "more of the same"? Programs tend to fall into two categories, those needing more memory than you have, and those which run easily in existing memory. AI, modeling, certain database programs, lots of things which could use a GB (or any other finite memory) might make use of 1GB memory. Editors, spreadsheets, communications, industrial control, graphics, compilation, CAD/CAM, are things which usually don't push the limit of current memory. Looking at accounting on some local workstations shows very few program which need more than 2MB of memory (even GNU emacs). If we are going to make good use of all that memory we will either need processors fast enough to drive many programs, or something better to do with all that memory. Of course I could mention that most people don't really *need* that much memory, and wouldn't use it at all, much less productively. Now that you're convinced that *you* need more memory, run vmstat for a day, using something like "vmstat 60 600 > /tmp/stat.log &" to get a reading every minute. Look at the free memory. If the machine is a workstation rather than being used for timesharing (many schools try to put 32 users on an 8MB Sun), the total memory in use is probably 4-12MB. Do most users need that in a workstation? I don't, as long as I have access to a large machine for those rare problems which can use that much memory. If a workstation is really going to have 1GB memory something better than "more of same" is going to be needed to justify the cost. -- bill davidsen (wedu@crd.GE.COM) {uunet | philabs}!steinmetz!crdos1!davidsen "Stupidity, like virtue, is its own reward" -me
bzs@bu-cs.BU.EDU (Barry Shein) (03/25/89)
> If a workstation is really going to have 1GB memory something better >than "more of same" is going to be needed to justify the cost. >-- > bill davidsen (wedu@crd.GE.COM) Well, to some extent, hoisted on my own petard, I've made similar arguments. But consider that 4 or 8MB workstations seemed just as out of sight a very few years ago (remember, the entire physical address space of an IBM370 architecture is/was [there are extensions now] 16MB! And that was the biggest mainframe architecture on the market.) Did changing over from 64K or 256K chips to get the larger memories significantly increase the price of the workstations? I don't think so, quite the opposite. It's probably safe to assume that when 4Mb/16Mb/64Mb chips come available over the next (very) few years they'll each be expensive for a short while and then drop down to the cost level of what they're replacing. So cost, other than replacing old equipment, will not be much of a factor. I don't think it will take too much to make you consider very large memories on your workstation. Just mapping in all/most of /bin and /usr/bin (etc) would probably make you want it if the guy next to you has that. It makes diskless/dataless workstations much more useful when you can use, say, 1/2 GB of memory as a read/write-through cache for the remote file system. Sure, cheap disks are going against that grain, but many still like the advantages of remote file systems, no noise, centralized backups and administration etc. And when large memories show up the disks will cease to be cheap, things are only cheap when the technology curves go out of kilter. The current memory rationing makes a 1GB disk for $5K seem very cheap. When you have a 1GB main memory on your workstation you'll need two or three of those just for swap space (?!) and we'll be right back where we started with our strategizing (unless something changes.) And don't talk to me about the bandwidth to get things in and out of those memories. And backups? Argh! As an analogy, who needs a 15MIPS workstation? Very few people, but they're available now and are cost competitive with slower workstations so, hey, that's what we all want. The others will wither on the vine. And trust the software tribes to eat up all available resources for you over time (have you seen the 3D window managers? etc) I do believe we will see a "Software Gap" in the near future with hardware vendors desparate for software applications which demand the new generation of hardware they're producing to induce people to upgrade. There's nothing more terrifying to hardware manufacturers than satisfied customers. -Barry Shein, Software Tool & Die
barnett@crdgw1.crd.ge.com (Bruce Barnett) (03/27/89)
Another thing to consider about 1 GByte Memory workstations, is that when the systems have more potential, the creative researcher finds a way to use that power. They thought 64K was enough. Then 256K was enough.... Bitmapped workstations revolutionized the way we work with computers. Suppose the workstation of the future had: Expert systems assiting you in creating new software, tapping into the knowledge base of the results of a million person-years of software experience. Hypertext encyclopedias available via USENET. Voice recognition systems, including personaility traits, inflections, etc. Artificial personalities. Real-Time, Real Colour 3D Imaging systems. When we worked with Punchcards, 64K was a lot. Video terminals 640K? Bitmapped graphics. 6.4M? Expert Systems 64M? ???? 640M? Give me enough memory, CPU power, tools, and time, and I would come up with one or two ideas. -- Bruce G. Barnett <barnett@crdgw1.ge.com> a.k.a. <barnett@[192.35.44.4]> uunet!steinmetz!barnett, <barnett@steinmetz.ge.com>
jfc@athena.mit.edu (John F Carr) (03/29/89)
In article <13433@steinmetz.ge.com> davidsen@crdos1.UUCP (bill davidsen) writes: >If the machine is a >workstation rather than being used for timesharing (many schools try to >put 32 users on an 8MB Sun), the total memory in use is probably 4-12MB. >Do most users need that in a workstation? Yes. At the moment, I am using about 27 Meg of virtual memory split between two workstations (4M & 6M RAM; 16 M swap). Processes: Saber C (a C interpreter running under X): ~7 M Emacs + subprocesses 2.5 M 2 large computational programs 2 M 4 pairs of (xterm+csh) 1.1 M X Server .7 M rrn+Pnews .5 M random small utilities, subshells (plus kernel & system processes) That is the static load; I also run compilers, the program I am working on, read mail, write files, etc... I find I can't fit all the programs I want to run into 16MB. I _don't_ have access to a large, fast machine for computation. Instead, I use X windows to run two workstations from a single display, and accept that overhead. -- John Carr "When they turn the pages of history, jfc@Athena.mit.edu When these days have passed long ago, bloom-beacon! Will they read of us with sadness athena.mit.edu!jfc For the seeds that we let grow?" --Neil Peart
consult@osiris.UUCP (Unix Consultation Mailbox ) (03/30/89)
In article <13433@steinmetz.ge.com> davidsen@crdos1.UUCP (bill davidsen) writes: > If the machine is a >workstation rather than being used for timesharing (many schools try to >put 32 users on an 8MB Sun), the total memory in use is probably 4-12MB. We have a pilot system running on a number of single-user diskless Sun 3/50s and I'll tell you exactly how much memory is in use on each of those workstations: the entire 4Mb. We had to double the size of all the server swap partitions just to keep the systems running. And even after taking the -g's and -gx's out of all the makefiles, *and* stripping all the executables, it's still Page City. >Do most users need that in a workstation? I don't, as long as I have >access to a large machine for those rare problems which can use that >much memory. I never needed more than the 4Mb in a 3/50 myself. Of course I was still doing most of my work on the Pyramids, which helps a lot. (They've all got >= 16M main memory and hundreds of Mb swap. Zippy!) phil