Slashdot Mirror
Interview With Original NT OS/2 Developers
leddhead writes "Was browsing over at the microsoft site when i ran across
this
interview with dave cutler of vms/winnt fame. It is interesting to
note how he stresses reliabilty over fancy graphics..." It's actually kind of an interesting interview if you ignore its PR-ish feel -- and the MS Word "?" problem if you're reading it in a Linux or Unix browser. The writer says the first NT OS/2 (NT's original name back in 1989) specs will be displayed at the Smithsonian soon. I wonder if this means Linus's first notes will be there someday. One can hope.
It's startling to look at their methodology- these guys are merging thousands of changes a night into builds, they hand it off to a very intense team of testers, who then test it for SIX HOURS??? ...hey, _wait_ a minute!"? It's really wild and kind of scary to consider that not only are they doing this, they still think it will work.
To put this into context, when the original Mac Finder, Teachtext, MacWrite etc. was being written, the coders would burn all night and then turn the program over to another computer program that would run _all_ _night_ making completely random and senseless GUI inputs all over to try and confuse and jockey the software into collapsing.
These Microsoft guys- it sounds like they are _manually_ trying out programs. For six hours. Do you have any idea how pathetically inadequate human input is to test such a program for six hours? How prone humans are to falling into patterns that don't cover all the inputs? How prone humans are to skipping _stupid_ inputs that might crash the machine?
This is a recipe for huge amounts of completely untested code to get out there. I'd suspected something like this, but reading Keepers Of The Build really drives it home forcibly- the project is TOO BIG to test. There is no way in hell they can jockey that software into all possible failure modes in six hours even _with_ 'virtual user' software- and how many sorts of machines are they testing on? Hell, even Apple ended up having to stop testing new software on every instance of machine they ever made- it got too expensive as there were hundreds of Macs and the logistics were impossible with so many software projects lining up to use the labs. It sounds like Microsoft is not even trying.
Instead they are doing things that _seem_ like they would be effective. They get people who _look_ really intense, they set up a combative situation so people will think 'Boy, they're really trying!'. They use the latest PCs (oh, but 'most powerful multiprocessor systems in the world', hell...) so people will think, 'Wow, they must really be able to debug much more than they could on _my_ machine!'. They are using a flatly ludicrous one-day cycle for the fastest builds, with the peculiar notion that they can track bugs better if the whole build is changing _faster_ than anybody else's development process... I presume when they can't debug it doing this, and new bugs keep happening faster than old ones leave (I bet NT 3.51 would have stood up to the w2000test.com load for longer than w2k), they presumably throw more programmers at the problem....
How many other people have read this seemingly impressive picture of their build situation and gone "...hey...
"For every 5 bugs that we squash, 7 more appear- so let's step up the pace and make the process happen five times as fast as it did! That'll help."
*shudder*
I guess logging out is a bit too complicated for me then. Because everytime I log out the box blue screens, not that it bothers me much since i only log out when I head home for the day, and for all I care the box could burst into flames at 5pm, as long as by 8am the next day it's magically put itself out and works. My problem with NT is that such simple operations can cause a crash. I don't know if the fault lies in the win32 api, or the kernel, or what. I really don't care. I've NEVER seen a unix system go completely belly up when a user logged out of it. Come to think of it, I've never seen any system do it before mine. And before you go blaming faulty hardware, we have two other machines in my department alone that exhibit the same behavior, then again they are the exact same machines using the exact same software so it could be a defective driver.
BTW, I'd prefer not to reevaluate my NT related knowledge, I'd rather just stop using it, wonder if I can convince my boss that my machine should be a linux box...
-matt
They said they got portability and extensibility...um, what about reliability, compatibility, and performance.
And why does it make me feel unsettled when a chief engineer is surprised and grateful that the operating system he just designed and coded actually *worked*?
I don't get what the big deal is. Are specs and design something new? Or just something new to Microsoft? I'm not an anti-MS troll, but this article really sounds like something fabulous happened. Wow...a decent design leads to a decent product, spec first, then code, premature optimization..., these don't sound amazingly novel to me...maybe they were in 89.
It's 10 PM. Do you know if you're un-American?
Just because the original OS/2 NT specs are being donated to the Smithsonian doesn't mean that they will be displayed. Most of the Smithsonian's treasures are stored in the basement and will never see the light of day. Many basement treasures are of questionable value since the Smithsonian will accept most any donation that has any chance, however remote, of someday being significant.
WINDOWS NT MUSEUM EXHIBIT DEBUT LESS THAN IMPRESSIVE
Exhibit "Looks Pretty, But Offers Little Substance"
WASHINGTON, DC (AP) - A new exhibit unveiled this week in the famed Smithsonian paid tribute to the Microsoft (NASDAQ: MSFT) operating system originally known as "NT OS/2." Currently known as "Windows 32/64 2005+ SP8R12", the operating system has also been known as "Windows NT", "Windows 2000", "Windows Memphis Moscow Beauregard 2001 Plus", "Windows For E-Commerce And Other Buzzwords", "Windows 16/32/64 PlusPack 312 With FrontPage Extensions", and "Steve". Although expectations were high among Smithsonian officials, the exhibit's debut was not without its problems.
"It was a real mess," said museum curator Steven Fleischmann. "The exhibit had only been open to the public for about twenty minutes when one of the curtains came crashing down, landing on some tourists from Guam. "Oh, it was awful," recounted one of the tourists. "The curtains fell, and after we got out from underneath them, all we could see was the wall that was behind the curtains. The wall was this mesmerizing deep blue color." Also painted on the wall was a series of alphanumeric characters used by the Smithsonian to track exhibits.
Then, one of the legs on the table that was holding the specs broke, and it sent the plastic case sliding on down to the floor." Security guards swarmed onto the scene to repair the table and re-hang the curtain, to minimize the amount of time that the exhibit was unavailable to the general public. "I was having a picnic with my kids," complained security guard Jeff Fenner. "It would have been nice if I could have fixed the problem without actually having to come here, but it's awfully damn hard to repair a table remotely." Fenner spoke on condition of anonymity.
Museum exhibit construction experts are blaming the embarassing incident partly on the exhibit's design. "Look at this," pointed out expert Louise Smith. "They've designed the exhibit so that the curtains are attached to the table. Apparently, the only reason they did this is to increase the visual attractiveness of the exhibit. But this is a dangerous design, as we found out this morning. If the curtains go down, it takes the table with it. I'm not sure that this is a model that should be imitated by future exhibits."
For its part, Microsoft is downplaying the exhibit fiasco. "Look," said an annoyed Ed Muth, "it's not our fault that the Smithsonian was unable to properly configure its table and curtains. If they had set things up correctly, the exhibit never would have gone down. In short, it's their fault, not ours. Our recommendation is to upgrade to a newer table and more durable curtains." Muth, a Microsoft project manager, also added a recommendation that the Smithsonian purchase "a large support contract from Microsoft."
Although Fleischmann remains optimistic about the exhibit's future, he still has some reservations. "Look at it," he said, gesturing. "It's very pretty. I think that people will want to look at it. I just have some very real concerns about the whole foundation of the thing, and I don't want to have to maintain an army of custodial staff to rescue the exhibit every time it collapses."
Nicholas Petreley contributed to this story.
We're going down, in a spiral to the ground
No actually it wasn't. I think Linus once said that it would never run on anything other than the Intel x86
It was designed so that all is needed to be ported is the micro-kernel on top of which is the rest of the OS
heh heh. Do a search for "Linux is obsolete". This is an old usenet thread from the early 90's where Linus argued against the micro-kernels with Andy Tanumbaum SP? (The author of Minix).
The irony as I see it is one of the merits of a Microkernel design is portability, yet Linux has become more widely ported than most microkernel OSes.
Of all the comments I've ever posted, this is definately one of them
Would there be any chance at all of a Slashdot-style (highest moderated questions win) interview with these guys. I for one would be very interested in their take of the OS market as it stands today and the technical merits and problems of NT and it's competitors (esp Unix, BeOS, Mac).
I use NT daily at work and (IMHO) it is very good most of the time. My problems mainly stem from what has been tacked onto NT (or, in MS speak, integrated). From what I hear about 2000, more integration is on the way...
MS gets bashed here all the time (often well deserved). But I know that they also happen to employ a lot of exceptionally talented people, especially in their R&D labs. Unfortunately, I've also heard that the minions charged with the task of translating spec into code often are not so talented. At a MS dev days event I attended a couple of years ago, the MS speaker said they suffer from the guy in a room syndrome- ie their projects too often depend on two or three exceptional people. At the time he was specifically talking about Excel.
Enough rambling...
EC
Nomethess doing a grep for .edu which is definitely the states with the .com and .gov and .org .net totaled up gives 143 email addresses, so thats less than half the credits file.
Allowing for a huge degree of error, which i imagine weighs against the concept that all the .coms and .orgs and .net are us, its looks pretty reliable that most of the contributers are not from the states.
Considering the states as the single biggest contributer, might be plausable however, but ultimately unprovable one way of the other, which is probably a good thing, as considering the issue is a pretty irrevelent thing to be doing anyway, a sense of community is what we should be striving for, but for the goal of pointing out that no one nationhood has any ownership of the kernel, it might be worth my while
A quick totaling up of some of the eu endings gives a total of 99 credits for .ie,.de,.uk,.fi,dk,.nl.,.fr,.se,.be,.it. Any how those germans love linux, they come in at a staggering 46 contributers compared to a measly runner up 16 for .uk, though wales did produce alan, so all is forgiven. Fair credit has to be given to the impressive showing for .nl with 13.
some guessword would make me suspect that there is a half to one credits per million inhabitants of a country, ahalf for big ones :-). germany is 80 million, nl is approx 12 (methinks), ie and .be are approx 3 ands o on.
More mad meanderings leaves me to predict that the states might have 100 legit members in the credits list :-)
I sometimes write stuff
Maybe she bites the head off of Microsoft in front of a carnival audience?
----
----
Open mind, insert foot.
written by G. Pascal Zachary. Amazon has it at:
1 7/qid=935594240/sr=1-37/002-8663731-751845 2
http://www.amazon.com/exec/obidos/ASIN/00293567
(On a totally unrelated note, is anybody else interested in taking up a collection to get the 'Dave' banner taken out of circulation? That guy gives me the creeps.)
As was pointed out earlier, the first thing everybody should do (if they're interested in the subject, of course) is to go find a copy of 'Showstopper,' written about the birth of NT. I found mine remaindered. You do get the anecdotes about Cutler's boorishness, but you also get a balanced look at the development of a modern operating system.
Now, I'm not enough of a kernel guy to argue about design specs, capabilities, etc., but after re-reading the book there are many things that pop to mind about Linux/NT:
0) Cutler's goals are/were initially very unix-like. When you get right down to it, NT in its initial design sounded (again, to unqualified ears) remarkably similar, with its ways of isolating the kernel, etc. I cannot imagine HOW much fun they have had working DirectX into NT, and I can't even begin to imagine how much Cutler hated it (based on his vision as seen in the interview and in showstopper.)
1) Cutler was adamant about getting rid of bugs. The surprising stability of W2k beta3 was frankly shocking to me, until I remembered that he was back in the fold. Yes, we are talking about a prime example of bloatware, but even so. Unfortunately, since the code is only reviewed by a limited number of people, they're always going to be behind the curve. Furthermore, Cutler is (probably) not responsible for the behavior of other programs from Microsoft, such as Exchange, SQL, etc., and I think that is from my experience the primary starting point for failure on NT boxes. Add to this the fact that there was again undoubtedly tremendous pressure on Cutler from Marketing to do those whizzy things that would compromise kernel stability, and I'm glad I don't have his job (or that of a programmer on Exchange.) I honestly don't see how bugs can be eradicated from the enduser experience. NT is much more stable than is posited here IF you can keep the number of Microsoft server apps to a minimum, preferably one. But things like Back Office pretend that you can have Exchange, SQL, and SMS all running on the same box- hell, I can't even get through installing such a combination without a crash. As long as Cutler is left alone and has sufficient authority, that *might* happen, but frankly I doubt it.
2) Portability is gone now for NT. I never did quite understand how Microsoft wouldn't pony up the money to keep NT alive on PPC chips, and I'm even more confused about Compaq shutting down Alpha NT support and development. Isn't having a valid counterweight to Intel more important than, say, 250 million to keep NT alive on those 2 platforms? Does Microsoft trust AMD to survive? In contrast, Linux is ported to gosh-knows how many machines already, and will continue to be. Seems to me keeping it alive on multiple platforms would be an investment for the future.
3) Graphics. The bane of Cutler's existence. I think had his crew been left alone to create a text mode only NT for v1, we would be looking at a totally different situation. Novell did just fine with text only screens, and Linux did too in the beginning. Trying to force NT to run before it was stable enough to walk was a mistake. It would have been better to layer it on later AFTER stability was worked out.
Ultimately, I think NT is doomed to failure against Linux- there are simply too many people using Linux now. But more, I think anything manmade is either made for money or art. And things made for art (or love, if you will) endure. Cutler has the artist in him, but his painting keeps getting smeared by people above him. Gates doesn't have art in him and never will- he may be a nerd, but he's no geek.
stored on computers from birth to the grave
I'm sure she's referring to her Microsoft stock. At least she can admit that Microsoft is history. (:
---
Funny. Linux was NOT designed with portability in mind, using a monolithic kernel. Yet by using a fairly bland set of assumptions about uProcessor design, it runs on nearly everything (didn't I just see something about it being on the dreamcast machines now???)
Meanwhile, NT runs on TWO platforms. and 64 bit NT will be a complete rewrite! Dear God man, if you can't even port to a 64 bit CPU, how transportable can your OS be?
--- It is not the things we do which we regret the most, but the things which we don't do.
The article does not reveal very much from the original spec, however I am curious how members of the Linux kernel development team feel about what was said. For instance, they mentioned not worrying about the memory footprint and the importance of sticking to the ideas of the few key people who write the spec. This seems somewhat different than the internet development of Linux. Any thoughts on the two different development models, and kernel planning in general?
The interview is a no-win situation. On the one hand, it would have been great to hear the view of a couple NT developers on how MS corporate culture opposes their stated goals of portability, etc. (as was hinted at). But such an interview wouldn't be permitted by the culture, which means it would have been a third-party interviewer, and the interview would have been posted somewhere other than www.microsoft.com. But it's unlikely that such a third-party interview would have been permitted............
So it's a shame.
Nothing worth doing is worth doing today.
The continual claim that NT only BSODs on "non approved hardware" is utter bunkum. Here we are running, or rather, attempting to run, Procomm on NT4 on brand new PIII Compaq Deskpros, with absolutely no additional hardware over the base spec.
Every time we try to run Procomm32, NT BSODs. Please note that I understand the difference between an application crash and an OS crash. Note also that I am explicitly stating that NT itself is crashing.
It seems that Procomm on NT4sp5 causes NT to die horribly. It's fine with sp4. Of course, sp4 isn't Y2K compliant, and as a financial institution we have to have ticks in all the Y2K boxes.
In any case, there's no way an app should be bringing down the OS. There's nothing dodgy or special about the hardware. There is no possible reason for this to happen, except that NT is basically not robust.
(I personally think NT3.5 was pretty robust; each subsequent release has been less so.)
There are many experienced Micros~1 engineers in the team here, some of them MCSEs. None of them has a good word to say about NT. They've had to make the long walk to the other side of the building too many times to fix a broken NT server. They just accept it as an unpleasant reality of life; NT crashes, but managers love NT and NT pays the bills.
"Portability, reliability, extensibility, compatability, performance."
Well, three out of... wait... no, no, no, two out of, no, wait...... Argh! Nevermind. :)
I got the same error... but kept trying and got in. Try it again, just like with Win95. ;)
Geeky modern art T-shirts