I've seen this before, but don't have a reference. There's a system that uses this approach to verify ports of number-crunching code. The code is run on two machines simultaneously, with the numeric results compared at key points. When they differ, you have a porting problem.
This is valuable for finding compiler and machine problems which affect the results of important numerical packages,
like NASTRAN.
I'll admit developers might be squirmish due to past history, but the point of the article was getting NEW developers. People who have yet to work with Apple and can't develop any of these distrusts.
3% market share and dropping. That's the reality, like it or not.
Developers don't trust Apple or Jobs
Apple, at various times, has told developers to use a whole series of technologies which they then abandoned. Copeland. OpenDoc. Rhapsody. X86 targets. Apple is hated for this.
Too many APIs Apple has too many APIs which do roughly similar things. Glueing NextStep, the MacOS, Objective-C, and UNIX together required too much duct tape.
There's also the suspicion that Apple's future is simple appliances like the iPod. The iPod is highly profitable; the desktop business is low profit. The iPod market is growing; the Apple desktop market is shrinking. At some point, economics (and Apple stockholders) may force a decision that Apple must concentrate on its entertainment business.
Collision detection for convex polyhedra is incredibly cheap when done right. Polygon soups are inherently more expensive, but there's been considerable progress in that area.
ODE is still at version 0.039, and not improving much.
Reliable dynamics simulators are hard. If you want one for a commercial game, there's Havok 2. All the free stuff is very limited.
It takes years of hard work to write a physics engine. If you're competent, in six months you'll have something that sort of works. From "sort of works" to "works" is years of effort. And it's not patches. It's theory. So the open source process doesn't work very well.
Good collision libraries are fun. I've written one,
as part of Falling Bodies. I think I was the first, back in 1996-1997, to use axis-oriented bounding boxes with GJK, which is what SOLID, and everybody else, uses now.
Lin and Canny are the ones who really cracked this problem. Before Lin and Canny, algorithms for collisions in a space with N objects with M faces each were O(N^2) * O(M^2). They got that down
to slightly worse than O(Nm), where Nm is the number of moving objects. Very clever.
I-Collide was the first generally available package for this. The original version was in LISP, which was translated to C, retaining much of the LISP style.
They used axis-oriented bounding boxes with a linear programming package. This had some problems with numerical error, and the linear programming package was rather bulky. But it demonstrated, back in 1996, what was possible.
Then everybody (well, the half dozen or so people into this stuff) went to work and built better systems.
Actually, collision detection is a pain to code, but well understood today. Collision response, the actual physics, is much harder.
The end result of all this is that games can now have really big worlds with working collision detection.
It's straightforward to user test software. What you need is a quiet room with a computer and some video gear. You need to be able to record someone at the computer, recording both the screen and the user's face and hands, in sync.
You bring in users and give them a few set tasks.
For an image processing program, one task might be "read in the images from this CD, edit out all telephone poles and wires in the background, then print the images on the printer." Each user gets about an hour or two worth of tasks.
You do this for five or so users.
Then you review the tapes. You look for every place where the user made a mistake, had to back up, became frustrated, encountered a bug, or could not make further progress. Any problem encountered by more than one tester must be fixed. Any problem that kept any tester from completing the task must be fixed.
After fixing the problem, you retest, using a mix of people from the previous round and new people.
It's worth editing the tapes down to a highlights reel of problems, and making the developers sit through it.
I can believe that this guy has a mailing list with 5 million E-mail addresses on it. That the recipients really, truly "want to receive" the mail, I would question. How was consent obtained? Implicitly, as part of some other transaction, or explicitly, with the explicit, verified consent of the recipient? One wonders.
Re:Is Ironport a black hat? MOD up parent please!
on
Spammer Sues SpamCop
·
· Score: 3, Interesting
Ironport might call it "legitimate e-mail marketing". But it's spam. Nobody buys Ironport products or services to send out their church newsletter. Unless you're a big-time bulker, you'd never be able to keep an Ironport A60 busy.
What's scary about Bonded Sender is that IronPort is putting a big hole in the anti-spam infrastructure. They have total control over this hole. They have no contractual relationship or legal obligation
to mail recipients that says they can't use it any way they want to. They just say "trust us". They can change the rules and open the spam floodgates once they have enough people relying on Bonded Sender.
We've heard that before.
They have a TrustE logo, but that's meaningless. All that means is that you agree to conform to your own privacy policy. Which you can change at any time.
They might, for example, change the rules to "conform to the four pillars of responsible E-mail" promoted by the Direct Marketing Association.
Those don't require double opt-in or a previous business relationship, so bulkers can trade mailing lists around. All they have to do is honor narrow opt-outs. (Opt out once for Viagra, once for mortgage refinancing, once for inkjet refills...)
I dropped SpamCop after they were taken over by IronPort. IronPort sells spamming engines. "Supports up to 10,000 simultaneous connections". "Can deliver up to 1 million messages per hour". "Removes constraints on outbound email marketing".
Yeah, yeah, there are "legitimate uses" for this thing. Right. Sure.
Even worse, they have a "Bonded Sender program, under which spammers pay a fee to Ironport to bypass spam filters. They charge a fee of $20 for each complaint, but allow one free complaint per million spams. They're vague about what a "complaint" is, and admit they don't use "AOL complaints". They may be counting only complaints that reach abuse@bondedsender.com.
Since they don't require that mail be marked as "approved by BondedSender", few people know how to complain. And they don't disclose their complaints, or who's in the "Bonded Sender" program.
They're trying hard to insure that all the major anti-spam systems are hardwired to let their spam through. They have patches for all the major spam detection programs.
The patches bypass all other spam checking if the source IP address has the DNS record that says it's listed with BondedSender. Now you understand why they bought SpamCop.
A useful check for mail programs is to check the BondedSender whitelist, then run a conservative Bayesian spam filter on the content. If BondedSender says it's not spam, but the spam filter says it is, ship it off to the BondedSender abuse address. Definitely do this for honeypots.
Any BondedSender mail that shows up at a honeypot should be reported on NANAE. That will help track how much, or how little, Ironport is really enforcing their rules.
Stanford has been doing badly with new science buildings.
First came the building for the IC designers. This has its very own wafer fab. Unfortunately, it's a 1980s wafer fab. Wafer fabs are too expensive to keep up to date unless they're heavily used. So it's obsolete. If you want something fabbed, you send it out.
Then came the Gates Computer Science building.
This is where AI went to die. Visit the second floor "Knowledge Systems Lab", and see all the empty cubicles with obsolete computers. The layout is wierd; the basement and first floor are difficult to get to from each other, and are connected only via a seminar room.
The corridors are devoted to a museum of old computer equipment, of mixed historical importance. ("Wow, an '80s DEC Ethernet hub! There's a Gandalf port selector box!")
Then there was the Lane Medical Library. Built with built-in stacks just as medical data was going online, it's used as office space. The immovable stacks remain, dividing up the space into long narrow office aisles.
Then came the New Engineering Quad. Finished at the height of the dot-com boom, it looks great. But it has all the wrong built-in stuff, like in-wall VCRs. More museum area. ("Wow, an original HP audio oscillator!") Good expresso bar, though.
The latest building is the Clarke Center for Biotech Engineering Stuff. Jim Clarke put up the money just before Netscape tanked. SGI contributed an obsolete supercomputer. Nobody seems quite sure what's going on there. The building has the overdesigned look of the free-spending dot-com era. There's steel and glass and balconies and atriums.
Microsoft operates a pulp fiction division.
Among other things, they publish "Spicy Air Tales". This started as a promo for an XBox game, but the stories are now being sold as paperback books. There's been talk of a movie deal.
The FTC press release is more informative than the Department of Justice press release.
(The CNN story is basically the DOJ press release.)
The DOJ press release says "The Lins have not been arrested at this time." The FTC is more explicit. They're wanted.
Arrest warrants are outstanding for defendants James Lin and Daniel J. Lin. In a criminal complaint issued by the U.S. Attorney's Office, these individuals have been charged with violations of the federal mail fraud laws as well as with criminal violations of the CAN-SPAM Act.
They sell Itanium-based servers. Nobody buys them.
(Well, actually they've sold a total of 13,000 Inanium CPUs. Not systems, CPUs. This is somewhere under 1% of the server market.) They sell MIPS servers running Irix. Nobody buys them either. SGI isn't even listed in the top five server vendors any more.
Almost all their buildings in Mountain View have been taken over by others.
Free-floating nanomachines will be severely energy-limited, like biology. Making something the size of a tree might take as long as it takes to make a tree. The power has to come from somewhere.
Assembly lines of nanomachines on IC-like substrates, supplied with external power, though, may actually be a useful manufacturing technology for small things.
I'm more worried about synthetic biology. So far, bioengineering has been a very crude trial and error process. Direct design of viruses and enzymes, let alone bigger organisms, doesn't work yet. But there's steady progress, and no reason it shouldn't work. That's going to mean designer diseases.
That big bump in March seems to have been SCO's announced stock buyback. So they blew a big chunk of their remaining cash, and the stock went back down anyway.
It doesn't get much worse than this:
Product sales are zilch.
Intellectual property sales are zilch.
Their VCs want their money back.
They're in litigation with four different Fortune 1000 companies. And not doing well.
Their CEO is widely viewed as an annoying loudmouth.
The stock is in a screaming dive.
They're widely hated.
The only upside is that there are no criminal charges or securities-fraud litigation.
Yet. That may come, if insiders enriched themselves during the stock runup.
Yes.
If automated memory management is to be retrofitted to C++, it should be reference-counted, not garbage collected. Calling destructors from the garbage collector means that they're called at more or less random times. This makes them unsuitable for controlling resources like files, windows, and such.
Perl has good reference-counting semantics, with strong and weak pointers. In structures with back-pointers, the back-pointers should be weak, which avoids cycles.
I've made some proposasl on this. But adding strictness to C++ isn't popular, which means another decade of buffer overflow exploits.
The GE 7HDL(TM) is powered by a 16-cylinder, twin turbocharged, electronically fuel-injected engine. This 6,250 ghp diesel engine is designed and manufactured for higher horsepower, greater fuel efficiency, lower emissions, improved reliability and easier maintenance. Elastic mounts isolate the engine and alternator from the platform to reduce platform-induced vibration This isolation reduces stress on components and creates a more comfortable environment for the driver.
The Nile delta floods every year, or at least it did before the Aswan Dam. I've always suspected that the Noah story came from somebody who'd seen the Nile flood, but didn't know this was a regular event.
There's a new book out about Krakatoa. Now that was an impressive flood. When the volcano blew up, the resulting wave carried a ship several miles inland. There are pictures. Nobody survived, though.
I've driven a Sunbeam Tiger, one of the classic overpowered cars. This was a conversion of a Sunbeam Alpine by Caroll Shelby, the originator of the Shelby Cobra. He put a Ford 4.2 liter V8 into a car that originally had a 4-cylinder 1.7 liter engine.
Fun. If you started off in third, you didn't notice any problem.
Slashdot Mirror
He supposedly made $18 million in 1997, and made the Forbes list of the top 40 richest entertainers.
This is valuable for finding compiler and machine problems which affect the results of important numerical packages, like NASTRAN.
There's a sucker born every minute.
There's also the suspicion that Apple's future is simple appliances like the iPod. The iPod is highly profitable; the desktop business is low profit. The iPod market is growing; the Apple desktop market is shrinking. At some point, economics (and Apple stockholders) may force a decision that Apple must concentrate on its entertainment business.
Collision detection for convex polyhedra is incredibly cheap when done right. Polygon soups are inherently more expensive, but there's been considerable progress in that area.
Reliable dynamics simulators are hard. If you want one for a commercial game, there's Havok 2. All the free stuff is very limited.
It takes years of hard work to write a physics engine. If you're competent, in six months you'll have something that sort of works. From "sort of works" to "works" is years of effort. And it's not patches. It's theory. So the open source process doesn't work very well.
Good collision libraries are fun. I've written one, as part of Falling Bodies. I think I was the first, back in 1996-1997, to use axis-oriented bounding boxes with GJK, which is what SOLID, and everybody else, uses now.
Lin and Canny are the ones who really cracked this problem. Before Lin and Canny, algorithms for collisions in a space with N objects with M faces each were O(N^2) * O(M^2). They got that down to slightly worse than O(Nm), where Nm is the number of moving objects. Very clever.
I-Collide was the first generally available package for this. The original version was in LISP, which was translated to C, retaining much of the LISP style. They used axis-oriented bounding boxes with a linear programming package. This had some problems with numerical error, and the linear programming package was rather bulky. But it demonstrated, back in 1996, what was possible. Then everybody (well, the half dozen or so people into this stuff) went to work and built better systems.
Actually, collision detection is a pain to code, but well understood today. Collision response, the actual physics, is much harder.
The end result of all this is that games can now have really big worlds with working collision detection.
You bring in users and give them a few set tasks. For an image processing program, one task might be "read in the images from this CD, edit out all telephone poles and wires in the background, then print the images on the printer." Each user gets about an hour or two worth of tasks. You do this for five or so users.
Then you review the tapes. You look for every place where the user made a mistake, had to back up, became frustrated, encountered a bug, or could not make further progress. Any problem encountered by more than one tester must be fixed. Any problem that kept any tester from completing the task must be fixed.
After fixing the problem, you retest, using a mix of people from the previous round and new people.
It's worth editing the tapes down to a highlights reel of problems, and making the developers sit through it.
That's how it's done. It's not rocket science.
It's been done. Ever seen a standalone X Terminal> Or the i-Opener. Both work reasonably well. Nobody wanted them.
I can believe that this guy has a mailing list with 5 million E-mail addresses on it. That the recipients really, truly "want to receive" the mail, I would question. How was consent obtained? Implicitly, as part of some other transaction, or explicitly, with the explicit, verified consent of the recipient? One wonders.
What's scary about Bonded Sender is that IronPort is putting a big hole in the anti-spam infrastructure. They have total control over this hole. They have no contractual relationship or legal obligation to mail recipients that says they can't use it any way they want to. They just say "trust us". They can change the rules and open the spam floodgates once they have enough people relying on Bonded Sender.
We've heard that before.
They have a TrustE logo, but that's meaningless. All that means is that you agree to conform to your own privacy policy. Which you can change at any time.
They might, for example, change the rules to "conform to the four pillars of responsible E-mail" promoted by the Direct Marketing Association. Those don't require double opt-in or a previous business relationship, so bulkers can trade mailing lists around. All they have to do is honor narrow opt-outs. (Opt out once for Viagra, once for mortgage refinancing, once for inkjet refills...)
Yeah, yeah, there are "legitimate uses" for this thing. Right. Sure.
Even worse, they have a "Bonded Sender program, under which spammers pay a fee to Ironport to bypass spam filters. They charge a fee of $20 for each complaint, but allow one free complaint per million spams. They're vague about what a "complaint" is, and admit they don't use "AOL complaints". They may be counting only complaints that reach abuse@bondedsender.com. Since they don't require that mail be marked as "approved by BondedSender", few people know how to complain. And they don't disclose their complaints, or who's in the "Bonded Sender" program.
They're trying hard to insure that all the major anti-spam systems are hardwired to let their spam through. They have patches for all the major spam detection programs. The patches bypass all other spam checking if the source IP address has the DNS record that says it's listed with BondedSender. Now you understand why they bought SpamCop.
A useful check for mail programs is to check the BondedSender whitelist, then run a conservative Bayesian spam filter on the content. If BondedSender says it's not spam, but the spam filter says it is, ship it off to the BondedSender abuse address. Definitely do this for honeypots. Any BondedSender mail that shows up at a honeypot should be reported on NANAE. That will help track how much, or how little, Ironport is really enforcing their rules.
First came the building for the IC designers. This has its very own wafer fab. Unfortunately, it's a 1980s wafer fab. Wafer fabs are too expensive to keep up to date unless they're heavily used. So it's obsolete. If you want something fabbed, you send it out.
Then came the Gates Computer Science building. This is where AI went to die. Visit the second floor "Knowledge Systems Lab", and see all the empty cubicles with obsolete computers. The layout is wierd; the basement and first floor are difficult to get to from each other, and are connected only via a seminar room. The corridors are devoted to a museum of old computer equipment, of mixed historical importance. ("Wow, an '80s DEC Ethernet hub! There's a Gandalf port selector box!")
Then there was the Lane Medical Library. Built with built-in stacks just as medical data was going online, it's used as office space. The immovable stacks remain, dividing up the space into long narrow office aisles.
Then came the New Engineering Quad. Finished at the height of the dot-com boom, it looks great. But it has all the wrong built-in stuff, like in-wall VCRs. More museum area. ("Wow, an original HP audio oscillator!") Good expresso bar, though.
The latest building is the Clarke Center for Biotech Engineering Stuff. Jim Clarke put up the money just before Netscape tanked. SGI contributed an obsolete supercomputer. Nobody seems quite sure what's going on there. The building has the overdesigned look of the free-spending dot-com era. There's steel and glass and balconies and atriums.
They never miss a revenue stream.
-
Arrest warrants are outstanding for defendants James Lin and Daniel J. Lin. In a criminal complaint issued by the U.S. Attorney's Office, these individuals have been charged with violations of the federal mail fraud laws as well as with criminal violations of the CAN-SPAM Act.
So if you know the whereabouts of those spammers, please contact the U.S. Attorney for the Eastern District of Michigan.The FTC also credits Spamhaus in assisting with the investigation.
Traditionally, the NYSE won't list a company with more than one class of stock. But they've softened that criterion in recent years.
They still sell workstations. Top of the line in desktops is a 2-CPU 800MHz RISC processor. Does anybody care?
They sell Itanium-based servers. Nobody buys them. (Well, actually they've sold a total of 13,000 Inanium CPUs. Not systems, CPUs. This is somewhere under 1% of the server market.) They sell MIPS servers running Irix. Nobody buys them either. SGI isn't even listed in the top five server vendors any more.
Almost all their buildings in Mountain View have been taken over by others.
Assembly lines of nanomachines on IC-like substrates, supplied with external power, though, may actually be a useful manufacturing technology for small things.
I'm more worried about synthetic biology. So far, bioengineering has been a very crude trial and error process. Direct design of viruses and enzymes, let alone bigger organisms, doesn't work yet. But there's steady progress, and no reason it shouldn't work. That's going to mean designer diseases.
That big bump in March seems to have been SCO's announced stock buyback. So they blew a big chunk of their remaining cash, and the stock went back down anyway.
It doesn't get much worse than this:
-
Product sales are zilch.
-
Intellectual property sales are zilch.
-
Their VCs want their money back.
-
They're in litigation with four different Fortune 1000 companies. And not doing well.
-
Their CEO is widely viewed as an annoying loudmouth.
-
The stock is in a screaming dive.
-
They're widely hated.
The only upside is that there are no criminal charges or securities-fraud litigation. Yet. That may come, if insiders enriched themselves during the stock runup.Perl has good reference-counting semantics, with strong and weak pointers. In structures with back-pointers, the back-pointers should be weak, which avoids cycles.
I've made some proposasl on this. But adding strictness to C++ isn't popular, which means another decade of buffer overflow exploits.
So there.
There's a prototype of the Cadillac Sixteen. The Veyron is also a prototype. In production next year, maybe.
There's a new book out about Krakatoa. Now that was an impressive flood. When the volcano blew up, the resulting wave carried a ship several miles inland. There are pictures. Nobody survived, though.
Fun. If you started off in third, you didn't notice any problem.
For that matter, Cadillac has a "1000 horsepower car." No way could it deliver that much power through the wheels, though.