cliffhanger@cup.portal.com (Cliff C Heyer) (09/14/89)
Re: Gregory/88000 vs 3081. >So what is a "workstation"? I have been unable to >make any real distinction between a "workstation" and a "high-performance PC". >The distinction (if there is one) is purely marketing hype. The distinction is purely I/O, *not* MIPS. All the 80386, 68030, etc. PCs are *not* running at full capacity due to memory & I/O subsystems being slower than the processor cycle time. Faster components blow the cost out of the water. Ask yourself why a 1974 DECsystem-10 could support 40 users with only 1 MIPS of processor power, and today we have PCs that have 8 MIPS but could NEVER handle 40 users with UNIX. The DECsystem-10 had a 20MB/s bus I/O BW, and had 2MB/s I/O to each of several MASSBUS disk drives. Today's PCs poke along with 5MB/s bus throughput and disk I/O of .2MB/sec (200KB/sec) even with "alleged" 1.5MB/sec SCSI. Try copying large files with your stopwatch for the proof. For example, on a (4 MIPS) VAXstation 3500 if I time the copy of a 1MB+ contiguous-sectored file from disk A to B, I get a rate of 1.3MB/second. (This is using SCSI w/large PC hard drives) That is: A=File size in millions of bytes. B=Stopwatch time to copy this file file from disk A to B. Throughput MB/sec = A/(B/2). (B is divided by 2 because B is the time to READ from A *and* write to B. Dividing by 2 gives the ballpark read or write time.) A VAXstation 3500 CPU w/box costs about $11K, and then you can hook up all the same SCSI stuff as the IBM PC. The cost is the same as a souped up 386, except the VAX CPU board goes for $10K instead of $3K. So you pay a *bit* more for that 1.3MB/sec. In fact, with the faster I/O you may get more work done with the 4 MIPS VAX than the Clipper, because the Clipper has memory wait states that slow it down to maybe 3 MIPS. Many 386 PCs give values of about only 200KB/sec which is not a "SCREAM" unless you compare to the PC XT or something (which is what manufacturers do to mislead you.) This is with SCSI too!!! One of the few exceptions is the Sun 386i which does a file read at 800KB/sec & write at 400KB/sec (however it was optimized for UNIX and does not run MSDOS too well). I have yet to see a PC that can SCREAM like a VAX. Just look at Byte. Almost EVERY benchmark they show for 1MB files is around 200KB/s for these machines. Try that on a VAX 11/780 and you'll get at least 800KB/s w/RP06s or 07s. Perhaps you could try this file copy on your machine and see what you get? I have heard of a few machines other than SUN that can do 750KB/sec. Use MSDOS to do this so there is no program overhead. If you only have one disk, you can copy to the NULL device(in which case don't divide B by 2). >We are/have been replacing our dependency on mainframe computing >by acquiring a network of dedicated, "high performance" (and relatively) >low cost "PC's". Some of these things are "workstations", but they all >use the AT style PC buss, and take too bloody long to do the disk i/o. Yup. And that's the way companies want it (& need to have it) If they all started to sell PCs today with 3MB/s to each disk and memory with those *non existent* PC profit margins, no one would buy their $1M mainframes in which case they would go broke. Also, to build 3MB/s to one disk on a PC would require expensive hardware which would raise the cost to $50,000+ per unit - making it no longer "cost effective" compared to the big iron. >It is cheaper (more effective) for us to have an Intergraph Clipper >chipset mounted in an AT-class PC take 8 hours to run our application than to >have an IBM-3081K do it in 3 and then send us back the results. If I remember, the 3081K (is it a 3081KX?) can give about 9 MIPS *per job*, first shipped in early '82, and had an internal BW of about 70MB/s. Your Clipper has more MIPS but no BW like 70MB/s. But the 70MB/s is divided up to about 3MB/s per disk channel and user. Even this is vastly faster than anything on the AT bus. >If I could have this processor (88000) in a machine with a decently >fast bus AND at a cost nearly the same, then it would be perfect. >...The VME bus based box would have >cost us THREE times the cost of the co-processor configuration! Fast components simply cost more, and it takes more time to engineer & test fast boards. Its the laws of physics. But what *you* can do is at least make sure you are getting state of the art for your buck. This means learning facts like the Amiga's ability to do 900KB/s DMA disk I/O, and not permitting a 33MHz 386 to be sold with 200KB/s without writing nasty letters to the editor of Infoworld, PC Week, etc. If you demand quality by making public the facts, the manufacturers will be forced to at least do the best they can. There are allot of ripoffs out there now because people have not educated themselves sufficiently to know what not to buy. I wish Infoworld and PC Magazine would include disk-to-disk copy benchmarks in all their tests. (but then I have not written to them either!) There is allegedly some kick-ass hardware out there in 386 AT land via SCSI/ESDI controllers ON BOARD bypassing the AT BUS. Mylex? ALR? I'm trying to find out before I buy. ONE thing is for sure - you can bet it won't come from IBM or DEC! They are too busy saving the BW for their *big iron*.
chris@mimsy.UUCP (Chris Torek) (09/15/89)
In article <22130@cup.portal.com> cliffhanger@cup.portal.com (Cliff C Heyer) writes: >... Try that on a VAX 11/780 and you'll get at least 800KB/s w/RP06s or 07s. Nah. I cannot speak for RP07s, but we have two RP06 washtubs plugged into an 11/785. You are lucky to get much over 500 KB/s, even doing full-track reads; put a reasonable file system on top of that and the read rate will drop, and the write rate will drop even more. (It is hard to find well-placed sectors on a full disk, and a 780's cpu does not help.) In contrast, a CDC Wren V on an async SCSI connected to a Sun 3/175 will typically move 500 KB/s through a Unix file system, even under the pokey old SunOS code. (The raw rate is about 1 MB/s, competitive with Eagles on an Emulex MASSBUS controller.) On synchronous SCSI, disks with real servo tracks and current densities will move 3 or 4 MB/s, if the bus at the other end of the SCSI adapter can handle it (and if the SCSI adapter itself can handle it). Cut that in about half for a Unix file system, unless the controller caches tracks (or, better, cylinders). (See the papers by Carson and Vongsathorn on adaptive disk rearrangement and other disk performance studies for some of the reasons for this.) -- In-Real-Life: Chris Torek, Univ of MD Comp Sci Dept (+1 301 454 7163) Domain: chris@mimsy.umd.edu Path: uunet!mimsy!chris
ggw@wolves.uucp (Gregory G. Woodbury) (09/16/89)
In article <22130@cup.portal.com> cliffhanger@cup.portal.com (Cliff C Heyer) writes: >Re: Gregory/88000 vs 3081. > >In article <mumble@wolves> ggw@wolves.uucp (Gregory G. Woodbury) writes: > >>So what is a "workstation"? I have been unable to >>make any real distinction between a "workstation" and a "high-performance PC". >>The distinction (if there is one) is purely marketing hype. Please allow me to correct this: "I have been unable to find any real distinction between SOME (so called) workstations and MOST (so called) high-performance PC's." The point of my article was to second the need to scream at the manufacturers in order to get a reasonably priced but fast buss. Its too bad that the EISA consortium seems to have buckled under and died. >Ask yourself why a 1974 DECsystem-10 could support 40 users with >only 1 MIPS of processor power, and today we have PCs that have 8 MIPS (or a 1979 PDP-11/60! with even less MIPS) >but could NEVER handle 40 users with UNIX. The DECsystem-10 had a >20MB/s bus I/O BW, and had 2MB/s I/O to each of several MASSBUS >disk drives. : : >A VAXstation 3500 CPU w/box costs about $11K, and then you can >hook up all the same SCSI stuff as the IBM PC. The cost is the >same as a souped up 386, except the VAX CPU board goes for >$10K instead of $3K. So you pay a *bit* more for that 1.3MB/sec. >In fact, with the faster I/O you may get more work done with >the 4 MIPS VAX than the Clipper, because the Clipper has memory wait >states that slow it down to maybe 3 MIPS. Actually, these co-processor implementations only slow down for the i/o - all universially (seem) to have an amount of their own dedicated memory that is NOT accessed via the AT-style buss. The Clipper board have 16MB of 1 wait dynamic RAM, which combined with the dual cache controllers and large chunk prefetches, make the things scream in terms of computations, but the I/O, being bottlenecked through the AT-buss and the host CPU, still s**ks. The 88K co-processors each have 20MB of similar DRAM. In fact, it is that we NEED the most memory available that boosts the costs of these machines up into the price ranges mentioned. With the recent stabilization of the memory market, the newest machines we bought are actually a little less expensive than machines bought 2 years or so ago. >Perhaps you could try this file copy on your machine and see >what you get? no need to benchmark, the disk i/o is directly limited by the hosting AT-class machine. What it can do in DOS, is the most that the co-processor can ever get. It may be a horrible way to handle the i/o, but it does make the machine reasonably easy to design and get to market. And that is what I am objecting to! Getting the 88K as a co-processor and placing it in a "high- performance" PC host (e.g. 33MHz '386 with 11MHz buss) costs significantly less than getting the 88K workstation from DG (for example). We knew that we could effectively use the 88K, but the budgets and things are such that we can finagle a less expensive co-processor based machine several times, whereas we cannot get the money together (all at once) to get the more expensive workstation. >>We are/have been replacing our dependency on mainframe computing >>by acquiring a network of dedicated, "high performance" (and relatively) >>low cost "PC's". Some of these things are "workstations", but they all >>use the AT style PC buss, and take too bloody long to do the disk i/o. > >Yup. And that's the way companies want it (& need to have it) If they >all started to sell PCs today with 3MB/s to each disk and memory with >those *non existent* PC profit margins, no one would buy their $1M >mainframes in which case they would go broke. (hmm, there are enough people in the business that there must be some kind of profit to make. Any capitalist knows that there has to be some kind of profit there to attract all the "fish". ;-) > >If I remember, the 3081K (is it a 3081KX?) can give about 9 MIPS *per >job*, first shipped in early '82, and had an internal BW of about 70MB/s. >Your Clipper has more MIPS but no BW like 70MB/s. But the 70MB/s is >divided up to about 3MB/s per disk channel and user. Even this is vastly >faster than anything on the AT bus. (I stipulate the advantages of "mainframe" io structure and channel architecture ;-) > >>If I could have this processor (88000) in a machine with a decently >>fast bus AND at a cost nearly the same, then it would be perfect. >>...The VME bus based box would have >>cost us THREE times the cost of the co-processor configuration! > >Fast components simply cost more, and it takes more time to engineer >& test fast boards. Its the laws of physics. But once they have been designed and tested, then they should cost somewhat less to produce and price should ramp down after the bus has been around for a while. The real price-killer of the VME system we were trying to configure was the ethernet controller! NatSemi only supported one or two possible cheapernet talkers and the cost of them was way out of sight. Even being willing to do some of the integration ourselves (by getting source and kernel hacking!) could not bring the cost of the system down to the level that we were willing to really consider. >But what *you* can do is at least make sure you are getting state >of the art for your buck. This means learning facts like the Amiga's >ability to do 900KB/s DMA disk I/O, and not permitting a 33MHz 386 >to be sold with 200KB/s without writing nasty letters to the editor >of Infoworld, PC Week, etc. If you demand quality by making public the >facts, the manufacturers will be forced to at least do the best they >can. There are allot of ripoffs out there now because people have not >educated themselves sufficiently to know what not to buy. I wish >Infoworld and PC Magazine would include disk-to-disk copy benchmarks >in all their tests. (but then I have not written to them either!) > >There is allegedly some kick-ass hardware out there in 386 AT land >via SCSI/ESDI controllers ON BOARD bypassing the AT BUS. Mylex? >ALR? I'm trying to find out before I buy. ONE thing is for sure - you >can bet it won't come from IBM or DEC! They are too busy saving the BW >for their *big iron*. Yes, they (IBM and DEC - and all the rest, including DG) will save the bandwidth and fast i/o for the *big iron* machines AND the high-end "workstations". This is just what the main problem is! The marketoids and management are so concerned in keeping the distinction between certain types of machine, and they manipulate the prices in order to do so, that they miss the boat. Actually, as we research the "next" machine we are going to get, some "workstations" are actually getting a real look. We are almost at the point of deciding that we have enough compute power for the moment, and we actually need a less number-cruncher of a machine in favor of a machine that can do efficient DMA i/o and network management and maybe acting as a file server. Sorry that this is so long, but this is of extreme interest to me and some of my colleagues. -- Gregory G. Woodbury Sysop/owner Wolves Den UNIX BBS, Durham NC UUCP: ...dukcds!wolves!ggw ...dukeac!wolves!ggw [use the maps!] Domain: ggw@cds.duke.edu ggw@ac.duke.edu ggw%wolves@ac.duke.edu Phone: +1 919 493 1998 (Home) +1 919 684 6126 (Work) [The line eater is a boojum snark! ] <standard disclaimers apply>
rfg@ics.uci.edu (Ronald Guilmette) (09/18/89)
In article <1989Sep16.044013.429@wolves.uucp> ggw@wolves.UUCP (Gregory G. Woodbury) writes: >In article <22130@cup.portal.com> cliffhanger@cup.portal.com (Cliff C Heyer) writes: >> >>>If I could have this processor (88000) in a machine with a decently >>>fast bus AND at a cost nearly the same, then it would be perfect. >>>...The VME bus based box would have >>>cost us THREE times the cost of the co-processor configuration! NOT TRUE! Add up *all* of the costs of a PC + co-processor and then check the price of an AViiON. >> >>Fast components simply cost more, and it takes more time to engineer >>& test fast boards. Its the laws of physics. > > But once they have been designed and tested, then they should >cost somewhat less to produce and price should ramp down after the >bus has been around for a while. Quite true. I think that DG's AViiON workstations are going to make many people reevaluate how they want to spend their $$$. The low end diskless node is < $8k. This is probably CHEAPER than a similarly equipped Compaq or whatever. They use the VME bus, I think. Onboard SCSI & ethernet also. >> >>There is allegedly some kick-ass hardware out there in 386 AT land >>via SCSI/ESDI controllers ON BOARD... ...and in workstation land! > Yes, they (IBM and DEC - and all the rest, including DG) will >save the bandwidth and fast i/o for the *big iron* machines AND the >high-end "workstations". This is just what the main problem is! The >marketoids and management are so concerned in keeping the distinction >between certain types of machine, and they manipulate the prices in order >to do so, that they miss the boat. That is probably a fair statement with respect to IBM and DEC, and is certainly a well known technique used for many years by IBM, but I think that it is very unfair to lump little DG in with those other massive monsters. One important thing to note. Unlike IBM and DEC, DG *lost* money last quarter. That is a *very* sobering experience for any organization and I'm sure that DG (as an organization) does *not* feel that they have the luxury of being able to try to play these marketing games (at least not right now). They know that they are going to have to battle their way into the workstation market and I believe that their very agressive price/ performance demonstrates this. > Sorry that this is so long, but this is of extreme interest >to me and some of my colleagues. You are not alone. // rfg
friedl@vsi.COM (Stephen J. Friedl) (09/18/89)
In article <1989Sep18.022722.16749@paris.ics.uci.edu>, rfg@ics.uci.edu (Ronald Guilmette) writes: > > Add up *all* of the costs of a PC + co-processor and then check > the price of an AViiON. For what it's worth, the AViiON is really an impressive box. We used to have a real ho-hum attitude to DG, but this new box is a real change for them. We get one in here in a month or two and are really hyped about it. Steve P.S. - I don't speak for V-Systems, no connection with DG, etc. -- Stephen J. Friedl / V-Systems, Inc. / Santa Ana, CA / +1 714 545 6442 3B2-kind-of-guy / {attmail uunet}!vsi!{bang!}friedl / friedl@vsi.com "This posting is a word to the wise, but you can read it too" - me
rec@dg.dg.com (Robert Cousins) (10/02/89)
In article <1989Sep16.044013.429@wolves.uucp> ggw@wolves.UUCP (Gregory G. Woodbury) writes: > Getting the 88K as a co-processor and placing it in a "high- >performance" PC host (e.g. 33MHz '386 with 11MHz buss) costs significantly >less than getting the 88K workstation from DG (for example). We knew that Actually, this is not the case. In any apples to apples comparison you will see not only that an equivalently configured machine DG machine WITHOUT the PC costs less. Furthermore, there is a performance penalty incurred when the PC is required to perform I/O operations (especially high bandwidth ones such as graphics). >we could effectively use the 88K, but the budgets and things are such that >we can finagle a less expensive co-processor based machine several times, >whereas we cannot get the money together (all at once) to get the more >expensive workstation. > Yes, they (IBM and DEC - and all the rest, including DG) will >save the bandwidth and fast i/o for the *big iron* machines AND the >high-end "workstations". This is just what the main problem is! The If you decide that $8000 is high end, then you are right, but frankly, I think your facts need to double checked. There are currently only 4 DG 88k workstations which all have approximately the same I/O bandwidth. While I will admit that it is quite fast, the I/O performance of the low end is almost the same as the high end. The major difference is in CPU speed. >marketoids and management are so concerned in keeping the distinction >between certain types of machine, and they manipulate the prices in order >to do so, that they miss the boat. > Actually, as we research the "next" machine we are going to get, >some "workstations" are actually getting a real look. We are almost at the >point of deciding that we have enough compute power for the moment, and >we actually need a less number-cruncher of a machine in favor of a machine >that can do efficient DMA i/o and network management and maybe acting >as a file server. See my paper in the September Computer Architecture News on this very subject. DMA is a must for performance at any level whenever you have more than one task running at a time. > Sorry that this is so long, but this is of extreme interest >to me and some of my colleagues. >-- >Gregory G. Woodbury >Sysop/owner Wolves Den UNIX BBS, Durham NC >UUCP: ...dukcds!wolves!ggw ...dukeac!wolves!ggw [use the maps!] >Domain: ggw@cds.duke.edu ggw@ac.duke.edu ggw%wolves@ac.duke.edu >Phone: +1 919 493 1998 (Home) +1 919 684 6126 (Work) Robert Cousins Dept Mgr, Workstation Dev't Data General Corp Speaking for myself alone. >[The line eater is a boojum snark! ] <standard disclaimers apply>