friedl@vsi.COM (Stephen J. Friedl) (09/23/89)
Hi folks, This appeared in Computer Systems News (18-Sep-89, p2), and part of it is reproduced here with permission only from my mom. #------------------------------------------------------- "AT&T Turns to Pyramid" - Basking Ridge, NJ AT&T is close to signing an OEM agreement with Pyramid Technology Corp. under which AT&T will offer Pyramid computers as high-end lures to attract IBM minicomputer users. For Pyramid, the deal would fit neatly into its strategy of expanding its business by selling its MIServer line of minicomputers and superminis through other vendors. The Pyramid system, which will carry an AT&T logo, is expected to replace the model 4000 in AT&T's 3B2 [sic] minicomputer line. [it's really the 3B4000]. An AT&T spokesman said he was unaware of a pending OEM deal with Pyramid, Mountain View, Calif. and Pyramid did not return calls by press time. Pyramid has told analysists it expects to soon sign an OEM agreement with a major computer company. [much more deleted] #------------------------------------------------------- I'm not a marketing type, but this looks like a positive development. We looked into a 3B4000 some time ago and got the pretty clear impression that they weren't doing so well, so maybe this was in fact the case. A question I have is whether they will do their own raw port of UNIX to it or rely on Pyramid's. Certainly the latter is the more expedient, but it opens them up to the same kind of nonportability problems they had with the UNIX PC where the operating systems aren't the same all around. Any comments? Steve Disclaimer: I don't work/speak for V-Systems, they don't know I'm posting this, 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
lyndon@auvax.uucp (Lyndon Nerenberg) (09/27/89)
friedl@vsi.COM (Stephen J. Friedl) writes: > I'm not a marketing type, but this looks like a positive >development. We looked into a 3B4000 some time ago and got the >pretty clear impression that they weren't doing so well, so maybe >this was in fact the case. This is in fact the case. The 3b4000 is a great machine for milking accounting data out of a 5ESS, but as a general purpose UNIX timeshare system it's hopeless! (I speak from experience here. We have a 3b4000 sitting in the corner of the machine room. AT&T said it would support 300-400 users. The president of the University believed them :-) The machine has been in production as a tape server for several months now. Performance is terrible.) [ I asked Ken Thompson why Plan 9 wasn't running on any AT&T hardware, "such as the 3b4000." His answer answer was most amusing :-) ] > A question I have is whether they will do their own raw port >of UNIX to it or rely on Pyramid's. Certainly the latter is the >more expedient, but it opens them up to the same kind of >nonportability problems they had with the UNIX PC where the >operating systems aren't the same all around. What I want to know is if AT&T is going to bugger the hardware sufficiently such that a UNIX distribution from Pyramid won't run on it. What would be *really* nice would be if AT&T would become a Sun OEM. We have ~$1M that we have to spend on AT&T products (credit for the 3b4000 we're sending back :-) I'd much rather buy a bunch of RISC servers and SPARCstations, or a big Pyramid. Mind you, the 3b2/1000's that we *are* getting look pretty impressive. [ Betcha never thought you would catch *me* saying *anything* good about AT&T :-) ] Lyndon Nerenberg VE6BBM / Computing Services / Athabasca University {alberta,decwrl,lsuc}!atha!lyndon || lyndon@cs.AthabascaU.CA "I think every man should have a wife. You can't blame everything on the government." -- Jed Clampett
gmb@occrsh.ATT.COM (Gary_M_Brammer) (09/27/89)
The reference to the AT&T 3B4000 as not a good time sharing system cannot be correctly stated without qualifying what has been done to tune the 3B4000. This machine straight out of the box will most likely be a poor performer in a general time share environment. It should not be referred to as a multi-processor machine, but loosely coupled multiple procesors or a tightly coupled network. We currently are using 3B4000's to run our production lines at the AT&T 3B/5ESS factory, and after correctly tuning the machines, we have experienced great results (100 users, 60000 TP1 type transactions per 8 hrs, and all processors > 60% idle). Every process executed on a 3B4000 should have some forethought about what I/O, memory resources, and CPU cycles are required (for example, vi should be marked to run on the processor associated with stdin and stdout). As far as dropping UNIX gurus' names, they are probably not the people that should be contacted for this type of question (ask them to write a book instead). The correct Bell Lab personnel to contact are located at Indian Hill BL in Naperville, IL. Gary Brammer Senior Engineer AT&T Oklahoma City Works
brian@ucsd.Edu (Brian Kantor) (09/28/89)
Good Grief! If the 3B4000 requires as much or more tuning as that to serve as a general purpose computer, I might as well buy an IBM 370 and heat the building with it as well. Perhaps in an environment where each terminal/user is going to do precisely ONE kind of application all the time every day, you could claim that a tuned machine is going to support lots of users. But in a university environment where the mix of tasks changes so wildly that no one can predict what will be running next, there is NO WAY to tune a machine to optimize performance of one kind of task without sacrificing some generality and the performance of all other kinds of tasks. What you need is raw frothing power. Or honest marketing that will tell me that my 42Q1000 isn't going to really be powerful enough to handle more than 8 to 10 users when 5 of them are using vi, three are running troff, and two are compiling modula programs. There's a big difference between a computer that can run the accounting department of a company and one that can support computer science students developing programs, starting with the way the thing handles job mixes and ending with the utilities provided. Computer manufacturers have to decide which market they're aiming for. More importantly, they have to tell their customers which environment their computers are optimized for and what its performance in that and other environments is going to be. Until that happens, you're going to have large hunks of iron that get returned to the vendor leaving lots and lots of ill-will behind them. Diogenes, where is your lamp? - Brian
gmb@occrsh.ATT.COM (Gary_M_Brammer) (09/29/89)
brian@ucsd.Edu (Brian Kantor) writes: >Good Grief! If the 3B4000 requires as much or more tuning as that to >serve as a general purpose computer, I might as well buy an IBM 370 and >heat the building with it as well. Although each processor on the 3B4000 has its own tuneable UNIX kernel, the tuning I was referring to was how resources were to be allocated to the processes being run (no process scheduler can predict the resources required by a program, it can only monitor current state/load of the system). >Perhaps in an environment where each terminal/user is going to do >precisely ONE kind of application all the time every day, you could >claim that a tuned machine is going to support lots of users. The machine I had previously referred to was running a large mix of applications including vi, troff, assorted compilers, multiple networks (TCP/IP, StarLan, RFS, Datakit VCS), and multiple relational database manager instances (some were remote databases). >But in a university environment where the mix of tasks changes so >wildly that no one can predict what will be running next, there is NO >WAY to tune a machine to optimize performance of one kind of task >without sacrificing some generality and the performance of all other >kinds of tasks. University computers typically run 5 applications: compilers, editors, USENET, games (rogue,hack,etc), and the one usual dork that insists on calculating PI to 2000 digits. :-) Anyway, applications can be categorized fairly easily (for example, wouldn't it be nice to be able to isolate game players so that they do not affect the response of "REAL" users). >What you need is raw frothing power. Or honest marketing that will >tell me that my 42Q1000 isn't going to really be powerful enough to >handle more than 8 to 10 users when 5 of them are using vi, three are >running troff, and two are compiling modula programs. Why would you trust an AT&T salescritter any more than another company's salescritter? We try to arm our sales people with the information you need to make an informed purchasing decision. Marketing "honesty" is at least partly the responsibility of the consumer to know at least something about what their requirements are (reasonable requirements, not the blue-sky that university professors think that they need). >There's a big difference between a computer that can run the accounting >department of a company and one that can support computer science >students developing programs, starting with the way the thing handles >job mixes and ending with the utilities provided. Where did I say we were doing accounting on the 3B4000? I am sorry that I was not more specific about what was being run on the 3B4000 when I said we "run our production lines" with it. >Computer manufacturers have to decide which market they're aiming for. >More importantly, they have to tell their customers which environment >their computers are optimized for and what its performance in that and >other environments is going to be. Until that happens, you're going to >have large hunks of iron that get returned to the vendor leaving lots >and lots of ill-will behind them. No argument with the above. However, computer systems are evolving into loosely coupled multi-processor networks and until users evolve past the mono-thinking/mono-processor stage, I fear that innovative machines like the 3B4000 will suffer the consequences of misunderstanding. >Diogenes, where is your lamp? > - Brian I think there is enough light for the world to make the RIGHT CHOICE. :-) Gary
lyndon@auvax.uucp (Lyndon Nerenberg) (10/02/89)
gmb@occrsh.ATT.COM (Gary_M_Brammer) writes: >We currently are using 3B4000's to run our >production lines at the AT&T 3B/5ESS factory, and after correctly tuning the >machines, we have experienced great results (100 users, 60000 TP1 type >transactions per 8 hrs, and all processors > 60% idle). I don't doubt this, however it sounds very much like you are running one (or a few) applications in a dedicated environment. This is not what I would consider a general purpose timesharing environment. > Every process >executed on a 3B4000 should have some forethought about what I/O, memory >resources, and CPU cycles are required (for example, vi should be marked >to run on the processor associated with stdin and stdout). Again, this is nice if you have a limited set of applications being run in a static environment. That's not a very good definition of a university computing environment. Besides, I had terminals spread across two ACP's. Where should I stamp vi to run in a situation like that. (And why should I have to stamp it in the first place? The kernel should be dealing with this issue.) >As far as dropping UNIX gurus' names, they are probably not the people >that should be contacted for this type of question (ask them to write a >book instead). The correct Bell Lab personnel to contact are located >at Indian Hill BL in Naperville, IL. It only took us six months to find out these people even existed, and another two months to get "permission" to talk to them. Of the bug reports we submitted, nearly 50% were rejected by AT&T out of hand. Of the remaining bugs, we received one fix. The remainder are still being looked into. This isn't what I would call responsive service ... I have a lot more faith into Thompson's comments than those I received from most of the support staff. Lyndon Nerenberg VE6BBM / Computing Services / Athabasca University {alberta,decwrl,lsuc}!atha!lyndon || lyndon@cs.AthabascaU.CA "I think every man should have a wife. You can't blame everything on the government." -- Jed Clampett
gmb@occrsh.ATT.COM (10/14/89)
We seem to be somewhat out of sync with our postings. However, I did respond to most of your comments on my second response to Brian Kantor at UCSD. As far as marking process to be executed on a specific processor is concerned, a program can be marked to execute on a class of processors with specific resources (such as STDIN and STDOUT, amount of free memory, attached file systems, etc.). The AT&T Education Centers offer excellent courses on performance tuning of the 3B4000, and this is also an excellent place to obtain the contacts necessary for "inside" information that makes your job easier. Also, USENET is a good place to ask (beg) for assistance. As frustrating to internal users of AT&T computers as it is, it must be "like totally" unbelieveable what outside users face. I am sorry that any customer was not able to obtain the help from AT&T required to bring a system into full production (see my previous comments about salescritters). The problem with a company as large as AT&T is finding the right person or group that can solve your problem. I personally know that there are experienced, knowledgeable support people located in Tier 4 support, but it always seems you must face the first level support that checks to see that you have booted the system. Gary Brammer Senior Engineer AT&T Oklahoma City Works