someone smarter than Dave Cutler
in [Device Drivers]
|
From: Jonathan de Boyne Pollard on 30 Mar 2010 09:06 > >>> >>> The big improvement that Daytona (NT 3.5) brought was a smaller >>> memory footprint because someone smarter than Cutler forced him to >>> allow it. >>> >> Cite? >> > _Showstopper_ mentions this. > I'm going to have to read that, someday. > Cutler had to be dragged kicking and screaming into allowing some > paging of the kernel. > I wonder how much of the stick that Dave Cutler gets is completely justified. I can see a few good reasons for not having a pageable kernel, not the least of which is that Microsoft is writing an operating system into which third-party device drivers are loaded. Others may not think of this as a development concern, but it certainly is to Microsoft. One of the things that Microsoft is, institutionally, very well aware of is the daft things that third party programmers do. It devotes a lot of time to accommodating daft programming practices, and at least some thought to how to avoid creating opportunities for further such time sinks. Making kernel mode pageable would have opened up a whole vista of new possibilities for third party device and filesystem driver programmers to do things wrongly. Maybe that was forseen. Certainly history has shown that it did indeed work out that way. If one hangs out in the various kernel-mode programming discussion fora, and reads the books, articles, and other documentation on the subject, one realizes how often "No, you may not do that at DISPATCH_LEVEL or higher." is the answer. The retrospective view is useful, too. The concern that a pageable kernel was addressing was how much RAM PC/AT compatibles generally had at the time. In 2004, Raymond Chen wrote an article noting that the pageable kernel prevents Windows NT from booting and running from a USB drive, something that people didn't want to do back in the early days of Windows NT but certainly want to do nowadays. One commenter observed in response that Linux, with a non-pageable kernel, can be booted and run from such a drive, observing: "Looks like Dave's conservative engineering design sense might have been right after all.".
From: Pat Farrell on 30 Mar 2010 10:49 Jonathan de Boyne Pollard wrote: >> _Showstopper_ mentions this. >> > I'm going to have to read that, someday. > >> Cutler had to be dragged kicking and screaming into allowing some >> paging of the kernel. >> > I wonder how much of the stick that Dave Cutler gets is completely > justified. IMHO, a lot. He had blinders on the topic of modern memory management. The Tenex/Tops-20 folks had it right, and they were all DEC. Now at the time, DEC was not really a single company, so a bit of the NIH spirit made sense from a corporate view. But his decisions hurt both VMS and NT, or rather his lack of experience in the state of the art. And in some cases, the start of the art we are talking about is 1969 art, implemented in the late 70s for VMS and late 80s for NT. > The retrospective view is useful, too. The concern that a pageable > kernel was addressing was how much RAM PC/AT compatibles generally had > at the time. When NT was in public beta test, it required 32MB of ram, miminal configuration. This was a problem, because in 1991 or so, most PC motherboards could not directly address more than 16MB, due to hardware constraints. The commodity PCs used a windowing scheme where memory was read from the disk into the bottom 16MB and moved up. Highly inefficient. Perhaps IBM's machines did it properly, but it was rare for generic PC/AT to do it well at all. Micron did it, but they were primarily a memory vendor making PCs to sell more memory. I bought a 486/66 Micron with SCSI everything and 32MB of ram specifically to beta test NT. -- Pat Farrell http://www.pfarrell.com/
From: Mark Crispin on 30 Mar 2010 12:40 On Tue, 30 Mar 2010, Pat Farrell posted: > The Tenex/Tops-20 folks had it right, and they were all DEC. Why is it that the closing theme from Camelot runs through my mind...about how in a halcyon time there was something fair and wonderful that is gone forever except in memory... > But his decisions hurt both VMS and NT, or rather his lack of experience > in the state of the art. And in some cases, the start of the art we are > talking about is 1969 art, implemented in the late 70s for VMS and late > 80s for NT. 1969-1970 seems to have been a critical turning point. It saw the release of Multics (although the project started in 1964), and the inception of Tenex (later TOPS-20) and UNIX. Also in that time, TOPS-10 and ITS, which were much earlier technology, had mid-life kickers in which you can draw a clear "before/after" line. I wonder if anyone will ever write an accurate history of that time, as opposed to the various bogus histories that overly tout self-promoters (Jobs, Gates, Stallman, Cutler, etc.) to the exclusion of others. Most of these so-called histories totally ignore the PDP-10, a platform that utterly dominated from the late 1960s until the late 1970s, and remained important until the late 1980s. And the notion that free software and software sharing didn't exist prior to the GNU religion is downright offensive. -- Mark -- http://panda.com/mrc Democracy is two wolves and a sheep deciding what to eat for lunch. Liberty is a well-armed sheep contesting the vote.
From: John Francis on 30 Mar 2010 12:56 In article <lcosn.20672$iu2.3907(a)newsfe15.iad>, Pat Farrell <pfarrell(a)pfarrell.com> wrote: >Jonathan de Boyne Pollard wrote: >>> _Showstopper_ mentions this. >>> >> I'm going to have to read that, someday. >> >>> Cutler had to be dragged kicking and screaming into allowing some >>> paging of the kernel. >>> >> I wonder how much of the stick that Dave Cutler gets is completely >> justified. > >IMHO, a lot. He had blinders on the topic of modern memory management. >The Tenex/Tops-20 folks had it right, and they were all DEC. Now at the >time, DEC was not really a single company, so a bit of the NIH spirit >made sense from a corporate view. > >But his decisions hurt both VMS and NT, or rather his lack of experience >in the state of the art. And in some cases, the start of the art we are >talking about is 1969 art, implemented in the late 70s for VMS and late >80s for NT. I don't believe it was ignorance as much as it was a deliberate choice. I always describe Cutler as having 20/20 tunnel vision - he knew what he considered to be most important, and focussed on getting that done. Other stuff that was just less important didn't get any attention. One thing people generally fail to mention is that Cutler delivered the initial versions of VMS (and, later, NT) reasonably close to the originally-targeted schedule; a feat which was practically unheard of with large software projects. Adding complicated memory management would have slowed down the schedule, so it got left off. I never worked directly with Dave Cutler, but I was working in the same group as him for a while (Languages & Tools; he was working on the C project, while I was working on the VMS debugger). I found him to be dogmatic, and convinced he was right 100% of the time - just like most software engineers. But over all I suspect he had a better accuracy rate than average. I also saw nothing to cause me to doubt the anecdotes telling of his arrogance and poor interpersonal skills.
From: Scott Lurndal on 30 Mar 2010 13:35
Mark Crispin <mrc(a)panda.com> writes: >I wonder if anyone will ever write an accurate history of that time, as >opposed to the various bogus histories that overly tout self-promoters >(Jobs, Gates, Stallman, Cutler, etc.) to the exclusion of others. Most of >these so-called histories totally ignore the PDP-10, a platform that >utterly dominated from the late 1960s until the late 1970s, and remained >important until the late 1980s. And the notion that free software and >software sharing didn't exist prior to the GNU religion is downright >offensive. You write "the PDP-10, a platform that utterly dominated from the late 1960's to the 1970's.". I assume you are speaking with your tongue in your cheek, since clearly the IBM 3[67]0 family and clones dominated the period in question, with the PDP-10 relegated to fourth or fifth tier after Burroughs, Sperry, Honeywell, Bull and CDC. scott |