CISC, RISC, and VLIW. It's very simple.
on
Emergence of SMT
·
· Score: 2
The bottleneck isn't inside the processor, we can clock multiply those suckers into the stratosphere. A modern Pentium or Athlon is going ten times the bus speed, and that can be increased easily. As long as you're executing out of L1 cache inside the processor, clock multipying is a big win and better design can go hang.
The BOTTLENECK is the memory bus. Refilling the cache from a hose that's 1/10th the speed of the processor. That's why CISC lasted so long in the first place, and is still with us today. CISC has variable length instructions. If you can express an instruction in 8 bits, you do so. 16 for the more complex ones, 32 bits for the really complex ones. So when you're sucking data into the cache 32 bits at a time, you can get 2 or 3 instructions in a 32 bit mouthful. (Or, in the case of pentiums, 64 bits to feed 2 cores, but the principle's the same.) You're optimizing for the real bottleneck with compressed instructions.
The fixed length instructions of RISC can be executed 2 at a time because you don't have to decode the first one to see where the second one starts. But Sparc, PowerPC, and even Alpha haven't displaced Intel because the real bottleneck is the memory bus, and bigger instructions aren't necessarily a win. (That and Intel translates Cisc to Risc inside the cache, and pipelines stuff.)
VLIW as iTanium picked it up sucks so badly because the real bottlneck is sucking data from main memory, and now they want 192 bits of it per clock! For only three instructions, and on average at least one will probably be a NOP. Crusoe has a MUCH better idea, sucking compressed CISC instructions in and converting them to VLIW in the cache (like Pentium and friends do for CISC to RISC).
This multi-threading stuff is just a way to keep the extra VLIW execution cores from being fed NOPs. They don't deal with the real problem, the memory bus de-optimization by reverting to full-sized instructions all the time.
Tim Berners-Lee's original public domain web browser code?
Saying free software development doesn't produce innovations is like saying the scientific method doesn't produce technological advancement. No, PEOPLE do. But sharing neat ideas is how they get turned into something useful rather than buried in a closet somewhere.
Texas Instruments invented the integrated circuit, but Intel was the one to make a microchip out of it. Atari didn't invent the videogame (both space war and pong were around before then), but Atari was the company that made the coin-operated video game industry happen. In each case, they got sued for patent infringment on things that they did the real work of making sucessful.
Giving other people explicit permission to improve on your ideas is what free software, open source, collaborative development, whatever you want to call it, is all about...
The first third of the book "The Innovator's Dilemma" by Clayton Christensen was all about the progress of hard drives from 14 inch washing machines to 2.5 inch laptop models.
Each switch to a smaller form factor (8 inch, 5 1/4 inch, 3.5 inch, etc) actually LOWERED the price/performance ratio and didn't seem to make sense, but it allowed the drive to be used in new situations (minicomputers for 8 inch, desktops for 5 1/4, early laptops for 3.5, modern laptops for 2.5.)
Who cares if the drive only has 5 gigs if it'll fit in your palm pilot?
Tell me again why anybody with a brain would voluntarily work for Microsoft?
They've had the permatemp thing, they've had the Executive Exodus (including Gates himself, if you want to look at it that way), and now they're the target of a discrimination lawsuit.
They can't even buy other companies to get people: http://www.pbs.org/cringely/pulpit/pulpit19990826. html
How did the mage-ette wind up with dirt all over her face, but perfect lipstick? And does she actually know more than two spells? (No magic missile, no burning hands...)
In the climax of the movie, the thief is up on the tower behind the two bad guys, they don't know he's there, and he does NOT do a backstab. No, he announces his presence so they can turn and kill him more easily.
What was the dwarf FOR? He didn't even get to hurt anybody in any of the fights. (A half-dozen melees, and nobody killed I could tell except that "spoons" guy...
I liked the ruby at the end with the disintegrate trap on it. "No thanks, I'll walk..."
So the elves hand over a magic sword to a race they dont' like "just because"? Would have been nice if they'd had REAL subtitles for the elvish whatserface was translating. Probably something like "It's cursed."
They all seriously needed a "ring of acting", "brasiers of acting", and maybe an ioun stone of +1 to acting... (I figured out what the curse on the red dragon sceptre was. Bad acting. Automatic failure of all acting rolls for the rest of the movie.)
>In fact, how many expensive Microsoft "Upgrades"
>have been mainly bug fixes? Windows 98, 98SE, and
>ME come to mind... All are essentially the SAME
>as Windows 95, only more stable (well, somewhat).
Cue yoda:
"Is the dark side stronger? No. Quicker. Easier to use. More seductive...."
As far as I can tell, 98 has more fonts, a version of explorer you can't uninstall, and "active desktop" which slows the entire system down by about a factor of eight. (Machines with 64 megs of ram should not constantly swap when you have only one window open.)
Let's see, active desktop, active directory, active X, even their marketing department is getting stuck in a rut...
Copyright covers the implementation, not the idea. If they want to get crotchety, then you can't use their text word for word, any more than you can copy a paragraph out of an encyclopedia when doing a book report in high school.
That said, bugtraq's just one more instance of third party support Microsoft is trying to do without by consuming. It's the black widow spider of software companies, and it's moving towards the extinction it deserves.
Reducing reporting of bugs won't reduce bugs. Quite the opposite. The answer is to stop using software that not only sucks, but has a company intent on making the whole experience less enjoyable.
Mars once had an atmosphere, water, the whole bit. Very similar to earth. Considering it was formed from the same general debris field as earth, this isn't really suprising.
What mars DIDN'T have was a strong magnetic field. Mars' geothermal core cooled (partly due to the planet's smaller size insulating it less, and possibly it may have less nuclear isotopes buried in it (again, proporitional to its mass) so it probably generated less heat in the first place.)
A magnetic field shields the planet from solar wind. Channels the charged particles around the planet, or down to smack head-on into the poles (forming the aurora borealis). Without a magnetic field, they strafe tangentially through the outer atmosphere, sandblasting it away. A molecule here, a few molecules there, over millions and eventually billions of years the atmosphere gets eroded away.
When the atmospheric pressure dropped too far, the oceans evaporated, probably freezing along the way as the planet lost its insulating blanket. The water vapor was just more atmosphere, to be sandblasted away by the solar wind.
Same thing would have happened to earth if we hadn't had the van allen belt and all protecting us. Planets farther out (such as the gas giants) don't have to deal with as intense a solar wind, and can rely on their own gravity to attract and retain gas faster than it gets stripped even without a magnetic shield. But they don't get much light or heat from the sun, either.
Linux will fragment when the PC fragments. The PC is commodity hardware, made from interchangeable parts acquired from different vendors according to known specifications and assembled by people who stamp a brand name on it. Linux is the exact same thing in software. The kernel, XFree86, GNU tools, raid freshmeat, integrate it into one big package.
Anybody who wants to can become a PC vendor, just like anybody who wants to can become a linux vendor. Same difference. It's not a cause of fragmentation, the little guys have to be MORE standard because they don't have the clout to push for changes in the standard base. Only by BEING standard can they get anybody to pay attention to them.
If Seagate made an incompatable hard drive that didn't conform to the ATA spec, Dell, Gateway and Compaq wouldn't use them. They'd fold. If Dell put out a computer that wasn't compatable with gateway and compaq's, they'd get bad PR and loose customers. The PC HAS fragmented before, and the offshoots died because the main base simply outgrew them and rendered them obsolete.
It's the exact same thing with linux. Compatability is evaluated by consumers and enforced by consumers who decide what they want to use. It's that simple. Tandy didn't make compatable stuff, they lost out. -IBM- stopped making compatable stuff (PS/2, PC Jr.), they lost out.
Any enhancement that can spread and be adopted by other vendors becomes a new standard. Any that can't diffuse in this way is eventually ostracized (even initially successful stuff like US Robotics HST modems: if it's single vendor proprietary it is DOOMED to inevitablly fall by the wayside. The commodity stuff out-evolves it over time. Guaranteed.)
We've got decades of history here, the trend's not hard to spot. Even for guys in suits.
> I think many of the MS-Windows instabilities are
> due to flawed design decisions. Not just code
> bugs.
I partially agree with this statement, but it's more often a case of NO design.
Take NT's migration from the OS/2 kernel, to a microkernel architecture, then hacking in ring 0 drivers for video and such because the performance sucked so badly, and then whatever they did to clean it up in 2000. Crisis juggling and hot potato management, pure and simple. "What shall we panic about today?" If it ain't broke, we've got a huge list of things to deal with first that ARE clearly broken. Plus marketing drives the agenda anyway.
> The Registry and it's access methods, for one.
Replace config.sys and autoexec.bat with something non-text based so clueless windows users don't think they understand it and trash things editing it. Cure far worse than the disease, but also another example of purely reactive design.
> Including foreign drivers into the NT4
> kernel for another.
I'll argue with the way they did it, but the idea of pluggable drivers IS something people have been pushing for even in Linux. (Can we say closed-source binary modules? They're evil, but it's usually better than not having them at all. No, I won't get sucked into an "incentive to reverse engineer" discussion, we're talking the old-style market microsoft's in.)
> FAT in general.
Fat came from DOS, which was originally architected for a machine with 64k of ram. For the OS and the programs combined! You couldn't AFFORD to do anything more fancy there: a linked list stored in a table, even hand coded in assembly, took up several precious kilobytes. After that, they were stuck with it (again, backwards compatability).
(Paul Allen wasn't really a slouch by the way, he saw Unix as the upgrade path for DOS until he retired in 1983 or so. THEN marketing took over.)
> Idle busyloop in 95|98|ME [how the EPA let that
> one slip, I'll never know].
That's implementation, pure and simple. Yes it was a bad decision to implement it that way, but how does that impact any other subsystems?
> Low tick frequency.
Also inherited from DOS, which they didn't want to change so as not to break application compatability with all the to-the-bare-metal programs that predated enough system resources to actually HAVE an operating system. (Although again, this is half an implementation issue. They could have emulated the slower one for applications...)
This is a configuration issue. Linux has "misc binary" support you can compile into the kernel, so the kernel can auto-detect the type of a binary and automatically run it via a support program.
Check the kernel documentation, or the help items under "make menuconfig"...
The fact that no distribution I know of currently configures this for you is probably mainly because they don't ship Wine yet, because it's still alpha code. (And because they tend not to ship java VMs due to licensing issues, although Kaffe is a candidate here...)
Wine allows you to run Win32 programs as native tasks under Linux. It does this by providing the Win32 API in Linux shared libraries, plus a few other tricks to actualy make it work (a server process to simulate some windows environment stuff and a loader that does some thunking on the executables it loads).
Running under Wine you're not actually running windows, just windows programs. You don't need a copy of windows installed, and you don't need a windows partition (loopback or otherwise).
So for things that CAN run under Wine, it's a better solution. They can show up seamlessly on your X desktop as normal X windows, and be launched by the kernel misc binary support straight from the command line or a gnome double-click.
The problem is, Wine can't run The Sims. And it's entirely possible it never WILL be able to run The Sims, because that game insists on loading a VxD in Ring 0 (for no apparent reason). Wine only emulates user mode code, not stuff that needs to run in ring 0 (I.E. wants to be part of the kernel.)
Wine's normal response to this is for the Wine developers to write their own implementation of common VxDs and include them in Wine, and recognize when a VxD is encountered and try to use their implementation instead. This helps with common stuff like DirectX, but doesn't help if the developers of the software actually DID write their own VxD. I don't know what the case is with The Sims, but on a theoretical level Wine can never be a 100% solution when the problem is inherently screwy. Windows allows programs to lobotomize its kernel on a whim via ring 0 VxD's. Linux ain't gonna do that. It's not that we're not actually able to, we're just not stupid enough to WANT to.
Plex86 doesn't care about VxD's. It lets them think they're running in Ring 0, although it's an emulated Ring 0. This means that Plex86 should be able to run The Sims without me having to reboot into a windows partion.
> Mmm-hmm. And figuring out what you can and can't
> do with GPL'ed software is like having to name
> each angel after you've counted it.
The GPL version 2 has been around for over a decade. It's only one license. You only have to figure it out once (and there's plenty of pages that walk you through it).
Microsoft changes their licenses on a monthly basis. Often retroactively.:)
> can the perimeter of the shape you form by
> folding ever exceed the perimeter of the
> original napkin?
Nope. You're doing two things:
1) A fold that makes reduces the permeter is hiding one or more corners and replacing each with a hypotenuse of the triangle, which is guaranteed to be smaller than the sum of the original two sides.
2) Any fold that increases perimiter is merely reversing the above process, trading in 'hypotenuse' for sections of previously hidden edge. (The "hypotenuse" here can be original edge, or edge created by folding. It still works out the same. You can't get more previously hidden edge than you started with (you can't reveal an edge that wasn't cached in a previous step), and you trade away the extra "hypotenuse" length you added in order to reveal those edges.)
>That is, Linux, BeOS and WinNT would all need
>different SMP drivers for the SMP Athlon boards
>to replace the APIC code.
Considering that Linux already supports SMP Alphas (up to 32-way has already been tested under 2.4-pre with no scalability problems), if Athlon uses the same technique I doubt there would be much new code required to get Linux to use it.:)
As with all apocalyptic visions, if this was easy to do the real world would have done it by now.
There's this little thing called entropy. Your nanowhatsises that are busily trying to reconfigure large quantities of molecular bonds in bulk: where do they get the energy to do so? If they're something capable of using the energy available in your body (metabolize sugars, etc), they're almost certainly something our immune system is designed to attack.
Even if they can get over the energy, thing, they're not going to be able to act TOO much faster than the reactions we're used to. The inherent chaos in the system (brownian motion) means molecules are whipping by darn fast, and everything at that level is twisting and jiggling. Standard reactions just grab and hold something and wait for the appropriate molecule to wander by and stick. They work because there's zillions of molecules wandering by, and if the reaction's to have any chance of working there's zillions of the appropriate type. When the right one hits, it sticks, and the various vaguely ionic attraction/repulsion forces (think magnetism, static cling, and the kind of constant vibration that turns sand into quicksand, all rolled into one) the molecule twists into a new shape (still twisting and bending and wobbling and jiggling, it just now spends the majority of its time in the new shape) and the reaction proceeds to the next step.
The reactions that ARE capable of proceeding rapidly aren't the kind that create more order. They create more DISORDER. Set fire to something. Dip it in acid. Blow it up. It's easy to rearrange molecular structure real fast, but the end result is scattered gasses and buckets of waste heat. Increasing order is a PAINFUL uphill climb, that's very slow and consumes a lot of energy.
We've had four and a half billion years of evolution fighting on this point. If there was a better way to do it that didn't have DARN obvious down sides, it would be the way it was done everywhere. Anything capable of taking over the planet in a week or two would have already DONE it at some point over the past few billions years.
There's a bunch of fun reactions we can't use locally. All sorts of exotic compounds that j ust so happen to explode on contact with water or oxidize amost immediately in our remarkably corrosive atmosphere. (Memo: rust ain't normal elsewhere in the universe. The life on this planet made the atmosphere that way a billion or so years after the fact (in part to kill off competing microorganisms that were poisoned by excess oxygen because their guts essentially rusted), and anything from elsewhere that was NOT used to a 20% oxygen atmosphere (where self-sustaining exothermic reactions can be set in motion and just continue! I.E. fire.)... It probably wouldn't live very long.)
>Motley Fool reports that Valenti believes that
>"Commerce trumpst the First Amendment."
Not Valenti. The -JUDGE-. And he DID say it (the actual quote was "is more important" rather than "trumps"). That's one of the grounds on which they asked him to recuse himself.
And to answer your title: high five it. The invisible hand is a good thing. Really.:)
> but what I'm getting at is the irony underlying
> the issues in that column. The same invisible
> hand that entertainment industry is trying to
> slap or shackle, is the one that also allowed
> the biggest entertainment mergers in history.
It's cyclical. Centralize for efficiency, decentralize to reduce bottlenecks. Good economic times favor large companies with economies of scale. Poor economic times favor small companies with low overhead.
The historical boom and bust cycle's gone all wonky due to the emergence of the internet, and the big companies have had a free ride for longer than normal. They can get incredibly arogant and inefficient and still do well because the overall economy's doing great.
Recessions normally take companies like this and break them across their knee. We just haven't had a recession in a while...
>But don't think this is going to be some huge >linux windfall, 'cause it ain't.
You're implying that there's enough left of SCO's user base that if every user moved over to Linux SIMULTANEOUSLY it would even register. This is simply not the case.
Linux has more seats than all the proprietary Unixen combined. That includes Sun's Solaris and legacy SunOS installs, IBM's AIX, HP-UX, Compaq Tru64...
A single large research institution or ISP switching from NT to Linux could have more of an impact than SCO's entire remaining installed base. The number of Linux boxes added to Beowulf clusters alone in any given month almost certainly buries SCO's entire installed base.
SCO is important for historical reasons, not installed base. They've been dying a slow death for a long time.
Their biggest impact these days is semantic: because of them Monterey is considered more than just the upcoing 64 bit version of AIX. IBM has invested enough in that project they'll probably ship it just to justify the work they've already done (possibly tax reasons, write off the expense), but I'd expect it to become instant legacy with an explicit transition path to Linux.
>The problem is that NT labor is readily >available, while Unix/Linux labor is not.
1) Maybe, but what are the colleges belching out today from their comp sci programs? Linux geeks. The OS is free, and all that source code is a great teaching tool, and the "I'd show you how it actually works, but then I'd have to kill you" mentality of Windows makes teaching the theory behind the hardware and software almost impossible.
What do geeks want to learn on their own? Linux. The best geeks are self-taught, and get a degree as an afterthought. The reason there aren't MORE Linux geeks is that until a few years ago only a select few had ever even looked at it. (Chicken and egg problem getting people to pay attention to something they've never seen in action.)
Looking at the deltas, both quality AND quantity are STRONGLY in Linux's favor. The backlog of professionally boot-camped NT admins is a historical artifact.
2) There's a REASON NT admins are so cheap. An MCSE doesn't mean you can think. Considering the test insists on the Microsoft answer to questions with more than one way to do it, it actually implies the opposite.
3) If you've admined an Irix box, Sun box, SCO box, or Linux 1.0, you've got the basic concepts down. There are no gratuitous changes, you don't have to be extensively retrained every couple years. How much good will NT 4 skills do on the W2K problem? (The certification certainly doesn't carry over.:) And they say Unix was fragmented...
I showed up as the show was dying and did what I could to save it. After talking with a lot of the vendors, I convinced the guy running it to change the time (stay open later), and bugged various guys from KLUG and KALUA to hold a LUG party there so vendors could see bodies. But it was too late, most of the vendors were gone by the time any changes could happen.
What went wrong:
1) The event ended at five. Most techies work day jobs. The event organizer said he was expecting people would only want to come to a Linux event on company time, and couldn't CONCEIVE of people wanting to come on their own initiative. (Sigh.)
2) The event didn't SCALE. It was the first event in the area, and was set up to only be successful if it had THOUSANDS of people. Hundreds showed up, but it was a failure. It had a HUGE auditorium, which looked empty even with a hundred people in it. It stretched over five days, so that if a thousand people showed up overall that would be about two hundred a day. Jeff from The Linux Show (www.thelinuxshow.com) told me a lot about making efficient use of space (narrow the aisles so it seems more crowded, if vendors leave put in tables and chairs where their booths were. Worst case scenario the boundaries are drapes on racks, so move them in so the empty space is walled off...) For their first outing, they needed an event that could be successful with 250 people but scale to at least ten times that. Instead they had one that would be a BOMB if less than 5000 people showed up.
The event was TOO LONG for what it had. They had Bruce Perens, Eric Raymond, Maddog, Emmet, but they were almost all gone by Wednesday night. If they'd had all of them for a two day event (friday and satuday, one day fo people coming on company time one day for the independents), it would have been a much more dense and compelling argument. And if everybody they had had showed up in half the time, they'd have had twice as many people at any given time.:)
3) There was advertising. The guy who ran it lost $100,000, and probably $30-40k of that was spent on radio and television ads. Targetted terribly. The vendors who came could have provided lists of names for direct mailings (postal, not spam, it's not as annoying because the had to spend money to do it.) But the vendors really weren't involved except to show up. Neither were the LUG. Both groups could see evil afoot, but expected the show to succeed on the sheer momentum of Linux. They thought "have a show, and they will come". Advertisers say you need to get the word out to potential customers seven times before they even remember you. The SAME potential customers. Many people "heard about it" but not enough times to actually bother making a DECISION. People procrastinate, you've got to remind them you exist a lot before they deal with you at all.
4) A show needs about a year of preparation and a large staff. The guy running it didn't delegate anything to anybody. There were no dress rehersals. There were almost no pre-confirmed attendees. (THIS is why you give big discounts to early registration.) Nobody was doing tapes of the conferences. There was no reception for the press the day before the show opened (forcing all the vendors to set up their booths ahead of time).
5) The place was almost impossible to find. Clear directions how to get there should have been on the website, and emailed to all confirmed attendees, plus posted on Linux Today at least. LOTS more signage was needed as well.
6) It wasn't targeted AT anybody. Corporate purchasing agents? Managers? Professional programmers? College students? Did it have a central theme? Who was the show FOR?
Finally, the people I talked to knew what was wrong (all of the above other people pointed out to me when I asked), but didn't feel involved in the process. I was some nut who drove up from Austin and I got more changed (fruitlessly as it turned out) than a lot of the people who were there, just by talking to people.
In the end, I grabbed a lot of boxes of magazines and CD's left behind in abandoned booths (freebies they were going to give out that would cost too muc to ship back) to take back to Austin with me. I've already given out about half of them, put them in the hands of actual Linux users. That's how I see this: an experience that we should salvage what we can from. In learning, and in resources.:)
P.S. One of Eric Raymond's writings is how Linux convention organizers can learn from the decades of experience of SF-Con organizers. ANybody thinking of putting on a con, go read eric's site. It's good. And email jeff at thelinuxshow.com who had more good suggestions about organization than any other person there. If you want to email me, bounce off my (now AMAZINGLY stale:) web site. My address is there.
>Also, it is questionable if the GPL would work. I >mean, IANAL, but my understanding is that the >copyright owner has the authority to license >their work for redistribution. However, Linux >doesn't appear to have a copyright owner. As a >result, if the GPL collapsed legally and needed >to be rewritten, then Linux is dead. Why? Because >NOBODY has the authority to redistribute the >kernel.
The GPL covers this eventuality, because it's version 2 "or future versions". The FSF could come out with a new version of the GPL that addressed whatever problem the license might have, and there you go.
Strangely enough, if the FSF got taken over it could come out with a new GPL that allows proprietary use, and kill it anyway, so really what are people losing by signing over the copyright? It's not like they're giving them more power than they already have: they can't.
That said, from an administrative standpoint requiring signing over the contract cripples GPL programs because the spur of the moment aspect of a lot of development is outright KILLED by it. If you have to fill out the equivalent of tax forms to get your contribution accepted, screw it.
This is why GNU languished for years and Linux took off immediately. Idealism gives us our frame of reference, but pragmatism is what works.
>To me, that's what free software should be about, >a way to facilitate the free sharing and communal >improvement of your code, not forcing others to >do the same if they don't want to.
Nobody's forcing anybody to use the SOFTWARE in the first place. People complaining about the coercive nature of the GPL are literally saying that although they got something for free, that's simply not good enough. Somehow, I can muster remarkably little sympathy with this point of view.
That said, using the LGPL is just as much the author's option as using the GPL (or a proprietary license, or a license that requires anybody wanting to make a copy to quack like a duck, for that matter...). Stallman's attempts to deprecate it are about as likely to restrict its use as Microsoft is likely to restrict Linux. It's out there, it's free, we can do what we want with it. Stallman has a certain amount of moral authority he can bring to bear, but that's about it.
Personally, I find the LGPL makes a lot more sense for libraries than the GPL does, by actually defining what IS a derived work and what isn't. It's a conceptually cleaner license in some ways.
Slashdot Mirror
The BOTTLENECK is the memory bus. Refilling the cache from a hose that's 1/10th the speed of the processor. That's why CISC lasted so long in the first place, and is still with us today. CISC has variable length instructions. If you can express an instruction in 8 bits, you do so. 16 for the more complex ones, 32 bits for the really complex ones. So when you're sucking data into the cache 32 bits at a time, you can get 2 or 3 instructions in a 32 bit mouthful. (Or, in the case of pentiums, 64 bits to feed 2 cores, but the principle's the same.) You're optimizing for the real bottleneck with compressed instructions.
The fixed length instructions of RISC can be executed 2 at a time because you don't have to decode the first one to see where the second one starts. But Sparc, PowerPC, and even Alpha haven't displaced Intel because the real bottleneck is the memory bus, and bigger instructions aren't necessarily a win. (That and Intel translates Cisc to Risc inside the cache, and pipelines stuff.)
VLIW as iTanium picked it up sucks so badly because the real bottlneck is sucking data from main memory, and now they want 192 bits of it per clock! For only three instructions, and on average at least one will probably be a NOP. Crusoe has a MUCH better idea, sucking compressed CISC instructions in and converting them to VLIW in the cache (like Pentium and friends do for CISC to RISC).
This multi-threading stuff is just a way to keep the extra VLIW execution cores from being fed NOPs. They don't deal with the real problem, the memory bus de-optimization by reverting to full-sized instructions all the time.
Rob
Saying free software development doesn't produce innovations is like saying the scientific method doesn't produce technological advancement. No, PEOPLE do. But sharing neat ideas is how they get turned into something useful rather than buried in a closet somewhere.
Texas Instruments invented the integrated circuit, but Intel was the one to make a microchip out of it. Atari didn't invent the videogame (both space war and pong were around before then), but Atari was the company that made the coin-operated video game industry happen. In each case, they got sued for patent infringment on things that they did the real work of making sucessful.
Giving other people explicit permission to improve on your ideas is what free software, open source, collaborative development, whatever you want to call it, is all about...
Rob
Each switch to a smaller form factor (8 inch, 5 1/4 inch, 3.5 inch, etc) actually LOWERED the price/performance ratio and didn't seem to make sense, but it allowed the drive to be used in new situations (minicomputers for 8 inch, desktops for 5 1/4, early laptops for 3.5, modern laptops for 2.5.)
Who cares if the drive only has 5 gigs if it'll fit in your palm pilot?
Rob
They've had the permatemp thing, they've had the Executive Exodus (including Gates himself, if you want to look at it that way), and now they're the target of a discrimination lawsuit.
They can't even buy other companies to get people: http://www.pbs.org/cringely/pulpit/pulpit19990826. html
Rob
In the climax of the movie, the thief is up on the tower behind the two bad guys, they don't know he's there, and he does NOT do a backstab. No, he announces his presence so they can turn and kill him more easily.
What was the dwarf FOR? He didn't even get to hurt anybody in any of the fights. (A half-dozen melees, and nobody killed I could tell except that "spoons" guy...
I liked the ruby at the end with the disintegrate trap on it. "No thanks, I'll walk..."
So the elves hand over a magic sword to a race they dont' like "just because"? Would have been nice if they'd had REAL subtitles for the elvish whatserface was translating. Probably something like "It's cursed."
They all seriously needed a "ring of acting", "brasiers of acting", and maybe an ioun stone of +1 to acting... (I figured out what the curse on the red dragon sceptre was. Bad acting. Automatic failure of all acting rolls for the rest of the movie.)
Rob
>have been mainly bug fixes? Windows 98, 98SE, and
>ME come to mind... All are essentially the SAME
>as Windows 95, only more stable (well, somewhat).
Cue yoda:
"Is the dark side stronger? No. Quicker. Easier to use. More seductive...." As far as I can tell, 98 has more fonts, a version of explorer you can't uninstall, and "active desktop" which slows the entire system down by about a factor of eight. (Machines with 64 megs of ram should not constantly swap when you have only one window open.)
Let's see, active desktop, active directory, active X, even their marketing department is getting stuck in a rut...
Rob
That said, bugtraq's just one more instance of third party support Microsoft is trying to do without by consuming. It's the black widow spider of software companies, and it's moving towards the extinction it deserves.
Reducing reporting of bugs won't reduce bugs. Quite the opposite. The answer is to stop using software that not only sucks, but has a company intent on making the whole experience less enjoyable.
Rob
What mars DIDN'T have was a strong magnetic field. Mars' geothermal core cooled (partly due to the planet's smaller size insulating it less, and possibly it may have less nuclear isotopes buried in it (again, proporitional to its mass) so it probably generated less heat in the first place.)
A magnetic field shields the planet from solar wind. Channels the charged particles around the planet, or down to smack head-on into the poles (forming the aurora borealis). Without a magnetic field, they strafe tangentially through the outer atmosphere, sandblasting it away. A molecule here, a few molecules there, over millions and eventually billions of years the atmosphere gets eroded away.
When the atmospheric pressure dropped too far, the oceans evaporated, probably freezing along the way as the planet lost its insulating blanket. The water vapor was just more atmosphere, to be sandblasted away by the solar wind.
Same thing would have happened to earth if we hadn't had the van allen belt and all protecting us. Planets farther out (such as the gas giants) don't have to deal with as intense a solar wind, and can rely on their own gravity to attract and retain gas faster than it gets stripped even without a magnetic shield. But they don't get much light or heat from the sun, either.
Rob
Anybody who wants to can become a PC vendor, just like anybody who wants to can become a linux vendor. Same difference. It's not a cause of fragmentation, the little guys have to be MORE standard because they don't have the clout to push for changes in the standard base. Only by BEING standard can they get anybody to pay attention to them.
If Seagate made an incompatable hard drive that didn't conform to the ATA spec, Dell, Gateway and Compaq wouldn't use them. They'd fold. If Dell put out a computer that wasn't compatable with gateway and compaq's, they'd get bad PR and loose customers. The PC HAS fragmented before, and the offshoots died because the main base simply outgrew them and rendered them obsolete.
It's the exact same thing with linux. Compatability is evaluated by consumers and enforced by consumers who decide what they want to use. It's that simple. Tandy didn't make compatable stuff, they lost out. -IBM- stopped making compatable stuff (PS/2, PC Jr.), they lost out.
Any enhancement that can spread and be adopted by other vendors becomes a new standard. Any that can't diffuse in this way is eventually ostracized (even initially successful stuff like US Robotics HST modems: if it's single vendor proprietary it is DOOMED to inevitablly fall by the wayside. The commodity stuff out-evolves it over time. Guaranteed.)
We've got decades of history here, the trend's not hard to spot. Even for guys in suits.
Rob
> due to flawed design decisions. Not just code
> bugs.
I partially agree with this statement, but it's more often a case of NO design.
Take NT's migration from the OS/2 kernel, to a microkernel architecture, then hacking in ring 0 drivers for video and such because the performance sucked so badly, and then whatever they did to clean it up in 2000. Crisis juggling and hot potato management, pure and simple. "What shall we panic about today?" If it ain't broke, we've got a huge list of things to deal with first that ARE clearly broken. Plus marketing drives the agenda anyway.
> The Registry and it's access methods, for one.
Replace config.sys and autoexec.bat with something non-text based so clueless windows users don't think they understand it and trash things editing it. Cure far worse than the disease, but also another example of purely reactive design.
> Including foreign drivers into the NT4
> kernel for another.
I'll argue with the way they did it, but the idea of pluggable drivers IS something people have been pushing for even in Linux. (Can we say closed-source binary modules? They're evil, but it's usually better than not having them at all. No, I won't get sucked into an "incentive to reverse engineer" discussion, we're talking the old-style market microsoft's in.)
> FAT in general.
Fat came from DOS, which was originally architected for a machine with 64k of ram. For the OS and the programs combined! You couldn't AFFORD to do anything more fancy there: a linked list stored in a table, even hand coded in assembly, took up several precious kilobytes. After that, they were stuck with it (again, backwards compatability).
(Paul Allen wasn't really a slouch by the way, he saw Unix as the upgrade path for DOS until he retired in 1983 or so. THEN marketing took over.)
> Idle busyloop in 95|98|ME [how the EPA let that
> one slip, I'll never know].
That's implementation, pure and simple. Yes it was a bad decision to implement it that way, but how does that impact any other subsystems?
> Low tick frequency.
Also inherited from DOS, which they didn't want to change so as not to break application compatability with all the to-the-bare-metal programs that predated enough system resources to actually HAVE an operating system. (Although again, this is half an implementation issue. They could have emulated the slower one for applications...)
Rob
Check the kernel documentation, or the help items under "make menuconfig"...
The fact that no distribution I know of currently configures this for you is probably mainly because they don't ship Wine yet, because it's still alpha code. (And because they tend not to ship java VMs due to licensing issues, although Kaffe is a candidate here...)
Rob
Running under Wine you're not actually running windows, just windows programs. You don't need a copy of windows installed, and you don't need a windows partition (loopback or otherwise).
So for things that CAN run under Wine, it's a better solution. They can show up seamlessly on your X desktop as normal X windows, and be launched by the kernel misc binary support straight from the command line or a gnome double-click.
The problem is, Wine can't run The Sims. And it's entirely possible it never WILL be able to run The Sims, because that game insists on loading a VxD in Ring 0 (for no apparent reason). Wine only emulates user mode code, not stuff that needs to run in ring 0 (I.E. wants to be part of the kernel.)
Wine's normal response to this is for the Wine developers to write their own implementation of common VxDs and include them in Wine, and recognize when a VxD is encountered and try to use their implementation instead. This helps with common stuff like DirectX, but doesn't help if the developers of the software actually DID write their own VxD. I don't know what the case is with The Sims, but on a theoretical level Wine can never be a 100% solution when the problem is inherently screwy. Windows allows programs to lobotomize its kernel on a whim via ring 0 VxD's. Linux ain't gonna do that. It's not that we're not actually able to, we're just not stupid enough to WANT to.
Plex86 doesn't care about VxD's. It lets them think they're running in Ring 0, although it's an emulated Ring 0. This means that Plex86 should be able to run The Sims without me having to reboot into a windows partion.
I like this.
Rob
> do with GPL'ed software is like having to name
> each angel after you've counted it.
The GPL version 2 has been around for over a decade. It's only one license. You only have to figure it out once (and there's plenty of pages that walk you through it).
Microsoft changes their licenses on a monthly basis. Often retroactively. :)
Rob
> folding ever exceed the perimeter of the
> original napkin?
Nope. You're doing two things:
1) A fold that makes reduces the permeter is hiding one or more corners and replacing each with a hypotenuse of the triangle, which is guaranteed to be smaller than the sum of the original two sides.
2) Any fold that increases perimiter is merely reversing the above process, trading in 'hypotenuse' for sections of previously hidden edge. (The "hypotenuse" here can be original edge, or edge created by folding. It still works out the same. You can't get more previously hidden edge than you started with (you can't reveal an edge that wasn't cached in a previous step), and you trade away the extra "hypotenuse" length you added in order to reveal those edges.)
No, this is not a mathematical proof. :)
Rob
>different SMP drivers for the SMP Athlon boards
>to replace the APIC code.
Considering that Linux already supports SMP Alphas (up to 32-way has already been tested under 2.4-pre with no scalability problems), if Athlon uses the same technique I doubt there would be much new code required to get Linux to use it. :)
Rob
There's this little thing called entropy. Your nanowhatsises that are busily trying to reconfigure large quantities of molecular bonds in bulk: where do they get the energy to do so? If they're something capable of using the energy available in your body (metabolize sugars, etc), they're almost certainly something our immune system is designed to attack.
Even if they can get over the energy, thing, they're not going to be able to act TOO much faster than the reactions we're used to. The inherent chaos in the system (brownian motion) means molecules are whipping by darn fast, and everything at that level is twisting and jiggling. Standard reactions just grab and hold something and wait for the appropriate molecule to wander by and stick. They work because there's zillions of molecules wandering by, and if the reaction's to have any chance of working there's zillions of the appropriate type. When the right one hits, it sticks, and the various vaguely ionic attraction/repulsion forces (think magnetism, static cling, and the kind of constant vibration that turns sand into quicksand, all rolled into one) the molecule twists into a new shape (still twisting and bending and wobbling and jiggling, it just now spends the majority of its time in the new shape) and the reaction proceeds to the next step.
The reactions that ARE capable of proceeding rapidly aren't the kind that create more order. They create more DISORDER. Set fire to something. Dip it in acid. Blow it up. It's easy to rearrange molecular structure real fast, but the end result is scattered gasses and buckets of waste heat. Increasing order is a PAINFUL uphill climb, that's very slow and consumes a lot of energy.
We've had four and a half billion years of evolution fighting on this point. If there was a better way to do it that didn't have DARN obvious down sides, it would be the way it was done everywhere. Anything capable of taking over the planet in a week or two would have already DONE it at some point over the past few billions years.
There's a bunch of fun reactions we can't use locally. All sorts of exotic compounds that j ust so happen to explode on contact with water or oxidize amost immediately in our remarkably corrosive atmosphere. (Memo: rust ain't normal elsewhere in the universe. The life on this planet made the atmosphere that way a billion or so years after the fact (in part to kill off competing microorganisms that were poisoned by excess oxygen because their guts essentially rusted), and anything from elsewhere that was NOT used to a 20% oxygen atmosphere (where self-sustaining exothermic reactions can be set in motion and just continue! I.E. fire.)... It probably wouldn't live very long.)
I'm not TOO worried about this. :)
Rob
Sigh...
Rob
>"Commerce trumpst the First Amendment."
Not Valenti. The -JUDGE-. And he DID say it (the actual quote was "is more important" rather than "trumps"). That's one of the grounds on which they asked him to recuse himself.
Rob
So do I. I wrote the column we're discussing. :)
And to answer your title: high five it. The invisible hand is a good thing. Really. :)
> but what I'm getting at is the irony underlying
> the issues in that column. The same invisible
> hand that entertainment industry is trying to
> slap or shackle, is the one that also allowed
> the biggest entertainment mergers in history.
It's cyclical. Centralize for efficiency, decentralize to reduce bottlenecks. Good economic times favor large companies with economies of scale. Poor economic times favor small companies with low overhead.
The historical boom and bust cycle's gone all wonky due to the emergence of the internet, and the big companies have had a free ride for longer than normal. They can get incredibly arogant and inefficient and still do well because the overall economy's doing great.
Recessions normally take companies like this and break them across their knee. We just haven't had a recession in a while...
Rob
http://www.wired.com/wired/archive/6.12/cringely.h tml
>linux windfall, 'cause it ain't.
You're implying that there's enough left of SCO's user base that if every user moved over to Linux SIMULTANEOUSLY it would even register. This is simply not the case.
Linux has more seats than all the proprietary Unixen combined. That includes Sun's Solaris and legacy SunOS installs, IBM's AIX, HP-UX, Compaq Tru64...
A single large research institution or ISP switching from NT to Linux could have more of an impact than SCO's entire remaining installed base. The number of Linux boxes added to Beowulf clusters alone in any given month almost certainly buries SCO's entire installed base.
SCO is important for historical reasons, not installed base. They've been dying a slow death for a long time.
Their biggest impact these days is semantic: because of them Monterey is considered more than just the upcoing 64 bit version of AIX. IBM has invested enough in that project they'll probably ship it just to justify the work they've already done (possibly tax reasons, write off the expense), but I'd expect it to become instant legacy with an explicit transition path to Linux.
Rob
>available, while Unix/Linux labor is not.
1) Maybe, but what are the colleges belching out today from their comp sci programs? Linux geeks. The OS is free, and all that source code is a great teaching tool, and the "I'd show you how it actually works, but then I'd have to kill you" mentality of Windows makes teaching the theory behind the hardware and software almost impossible.
What do geeks want to learn on their own? Linux. The best geeks are self-taught, and get a degree as an afterthought. The reason there aren't MORE Linux geeks is that until a few years ago only a select few had ever even looked at it. (Chicken and egg problem getting people to pay attention to something they've never seen in action.)
Looking at the deltas, both quality AND quantity are STRONGLY in Linux's favor. The backlog of professionally boot-camped NT admins is a historical artifact.
2) There's a REASON NT admins are so cheap. An MCSE doesn't mean you can think. Considering the test insists on the Microsoft answer to questions with more than one way to do it, it actually implies the opposite.
3) If you've admined an Irix box, Sun box, SCO box, or Linux 1.0, you've got the basic concepts down. There are no gratuitous changes, you don't have to be extensively retrained every couple years. How much good will NT 4 skills do on the W2K problem? (The certification certainly doesn't carry over. :) And they say Unix was fragmented...
Rob
What went wrong:
1) The event ended at five. Most techies work day jobs. The event organizer said he was expecting people would only want to come to a Linux event on company time, and couldn't CONCEIVE of people wanting to come on their own initiative. (Sigh.)
2) The event didn't SCALE. It was the first event in the area, and was set up to only be successful if it had THOUSANDS of people. Hundreds showed up, but it was a failure. It had a HUGE auditorium, which looked empty even with a hundred people in it. It stretched over five days, so that if a thousand people showed up overall that would be about two hundred a day. Jeff from The Linux Show (www.thelinuxshow.com) told me a lot about making efficient use of space (narrow the aisles so it seems more crowded, if vendors leave put in tables and chairs where their booths were. Worst case scenario the boundaries are drapes on racks, so move them in so the empty space is walled off...) For their first outing, they needed an event that could be successful with 250 people but scale to at least ten times that. Instead they had one that would be a BOMB if less than 5000 people showed up.
The event was TOO LONG for what it had. They had Bruce Perens, Eric Raymond, Maddog, Emmet, but they were almost all gone by Wednesday night. If they'd had all of them for a two day event (friday and satuday, one day fo people coming on company time one day for the independents), it would have been a much more dense and compelling argument. And if everybody they had had showed up in half the time, they'd have had twice as many people at any given time. :)
3) There was advertising. The guy who ran it lost $100,000, and probably $30-40k of that was spent on radio and television ads. Targetted terribly. The vendors who came could have provided lists of names for direct mailings (postal, not spam, it's not as annoying because the had to spend money to do it.) But the vendors really weren't involved except to show up. Neither were the LUG. Both groups could see evil afoot, but expected the show to succeed on the sheer momentum of Linux. They thought "have a show, and they will come". Advertisers say you need to get the word out to potential customers seven times before they even remember you. The SAME potential customers. Many people "heard about it" but not enough times to actually bother making a DECISION. People procrastinate, you've got to remind them you exist a lot before they deal with you at all.
4) A show needs about a year of preparation and a large staff. The guy running it didn't delegate anything to anybody. There were no dress rehersals. There were almost no pre-confirmed attendees. (THIS is why you give big discounts to early registration.) Nobody was doing tapes of the conferences. There was no reception for the press the day before the show opened (forcing all the vendors to set up their booths ahead of time).
5) The place was almost impossible to find. Clear directions how to get there should have been on the website, and emailed to all confirmed attendees, plus posted on Linux Today at least. LOTS more signage was needed as well.
6) It wasn't targeted AT anybody. Corporate purchasing agents? Managers? Professional programmers? College students? Did it have a central theme? Who was the show FOR?
Finally, the people I talked to knew what was wrong (all of the above other people pointed out to me when I asked), but didn't feel involved in the process. I was some nut who drove up from Austin and I got more changed (fruitlessly as it turned out) than a lot of the people who were there, just by talking to people.
In the end, I grabbed a lot of boxes of magazines and CD's left behind in abandoned booths (freebies they were going to give out that would cost too muc to ship back) to take back to Austin with me. I've already given out about half of them, put them in the hands of actual Linux users. That's how I see this: an experience that we should salvage what we can from. In learning, and in resources. :)
P.S. One of Eric Raymond's writings is how Linux convention organizers can learn from the decades of experience of SF-Con organizers. ANybody thinking of putting on a con, go read eric's site. It's good. And email jeff at thelinuxshow.com who had more good suggestions about organization than any other person there. If you want to email me, bounce off my (now AMAZINGLY stale :) web site. My address is there.
Rob
>mean, IANAL, but my understanding is that the
>copyright owner has the authority to license
>their work for redistribution. However, Linux
>doesn't appear to have a copyright owner. As a
>result, if the GPL collapsed legally and needed
>to be rewritten, then Linux is dead. Why? Because
>NOBODY has the authority to redistribute the
>kernel.
The GPL covers this eventuality, because it's version 2 "or future versions". The FSF could come out with a new version of the GPL that addressed whatever problem the license might have, and there you go.
Strangely enough, if the FSF got taken over it could come out with a new GPL that allows proprietary use, and kill it anyway, so really what are people losing by signing over the copyright? It's not like they're giving them more power than they already have: they can't.
That said, from an administrative standpoint requiring signing over the contract cripples GPL programs because the spur of the moment aspect of a lot of development is outright KILLED by it. If you have to fill out the equivalent of tax forms to get your contribution accepted, screw it.
This is why GNU languished for years and Linux took off immediately. Idealism gives us our frame of reference, but pragmatism is what works.
Rob
>a way to facilitate the free sharing and communal
>improvement of your code, not forcing others to
>do the same if they don't want to.
Nobody's forcing anybody to use the SOFTWARE in the first place. People complaining about the coercive nature of the GPL are literally saying that although they got something for free, that's simply not good enough. Somehow, I can muster remarkably little sympathy with this point of view.
That said, using the LGPL is just as much the author's option as using the GPL (or a proprietary license, or a license that requires anybody wanting to make a copy to quack like a duck, for that matter...). Stallman's attempts to deprecate it are about as likely to restrict its use as Microsoft is likely to restrict Linux. It's out there, it's free, we can do what we want with it. Stallman has a certain amount of moral authority he can bring to bear, but that's about it.
Personally, I find the LGPL makes a lot more sense for libraries than the GPL does, by actually defining what IS a derived work and what isn't. It's a conceptually cleaner license in some ways.
Rob