For a computational number-theory project I run (PIES) I was doing some preparatory number-crunching work before handing out jobs to clients.
My code is reasonably-optimised portable C, and I set it running on a 533MHz 21164 (from 1996, IIRC?), and a 2.8GHz P4.
They were running at within 5% of the speed of each other. The P4 architecure, unless you diddle in assembly, is pants. (Yes, I compiled using Intel's compiler.)
And given that this Alpha machine has been loaded >1.0 since time began - it's unbelievably rock solid: ecmnet@megaspaz:/etc/apache$ uptime
21:26:25 up 507 days, 8:41, 20 users, load average: 1.00, 0.96, 0.92
In theory, being from the kind of technical background that I am, I ought to fawn over every column, but, to be honest, I find his usual statements to be a bit feeble, a gentle puff, with no real gusto. He does pull his punches.
Normally.
However, this one has broken that mould. There were no punches pulled, and he completely nailed his colours to the mast. Good on him.
However, I'd be tempted to say that he's even made himself a target of Microsoft lawyers, as he has made allegations which could be, if false, be taken as libelous (or otherwise defamatory). (Not that I believe they are false.)
Will the posse of lethal attack-lawyers be set on him for it? Or will MS just hope it gets forgotten about as quickly as possible?
""" Some people spend hours every day triaging bugs. Bugzilla is a workflow miracle in the sense that it allows this volume of bugs to be dealt with effectively. """
I think you meant "Mozilla is a freaking disaster area if it allows this volume of bugs to exist in the first place".
I'm a vocifereous pro-OSS advocate. Yet even I admit that Mozilla's a steaming pile, an embarassment to the community as a whole.
""" Consider what happens when you do a new/malloc: a) the allocator wanders through some lists looking for a slot of the right size, then returns you a pointer. b) This pointer is pointing to some pretty random place.
With GC, a) the allocator doesn't need to look for memory, it knows where it is, b) the memory it returns is adjacent to the last bit of memory you requested. The wandering around part happens not all the time but only at garbage collection. And then (depending on the GC algorithm) things get moved of course as well. """
The guy knows nothing about _either_ efficient dynamic memory allocators, nor garbage collectors, he has no right to pass comment on them.
Why couldn't your post be nearer the top? I was nearly giving up on this story due to the unbelievable amount of misinformation and general gibbering!
I couldn't agree more with your points, the one thing that stood out most glaringly in the article was the one you pick on too: """ One of the key breakthroughs is an "intermediate representation," a kind of lingua franca that gives the software the flexibility to translate from one platform to another. """
So one of the breakthroughs is to use a bytecode in the same way that Java has since day 1. That's so not a breakthrough. The breakthrough is to get it _fast_. The claimed 80% is very impressive, unless that's emulating x86/windows on x86/linux, in which case it's crap.
"Calculating Pi to a billion digits is not floating point math, "
I call bullshit.
George Woltmann, for his bignum multiplications in GIMPS, uses Complex (i.e. FP) FFTs. Yves Gallot, for his bignum multiplications in Proth, uses Complex (i.e. FP) FFTs. Phil Carmody (i.e. me) for his bignum multiplications in ForEis, uses Complex (i.e. FP) FFTs.
Given that Xavier Gourdon hasn't realeased the source to PiFast, we can't know for sure, but given his comments in section 2.2 at http://numbers.computation.free.fr/Constants/A lgor ithms/fft.html wouldn't you be likely to infer that Xavier Gourdon for his bignum multiplications in PiFast, uses Complex (i.e. FP) FFTs?
The point that sems to have not been made is that when you plug in a 487 "coprocessor", it actually takes over the whole system, and your old 486SX is pretty much redundant. i.e. the 487 is a whole 486DX.
Primegen comes with a 'count' function which does no output, that would entirely isolate the CPU-bound component.
However, chosing to not turn on any optimisation for some of those tests means that these benchmarks were the most "synthetic" I've every seen -- they certainly corresponded to nothing real.
I don't know what to believe in that article - it is quite poorly researched: " Unfortunately, the pins are rather large to be electrical and mechanical devices. Only a few hundred fit on a package that contains a processor with several million transistors. Bandwidth, thus, is constrained. "
A few hundred? Explain why IBM manage to get more than five _thousand_ pins on their POWER packages.
This way, when "the enemy" notice weird TCP connection attempts to your box, they'll know that you're using port-knocking, and decide to hammer every single port in random order just on the offchance.
It's a great way of finding out if your TCP stack has memory leaks.
This is nothing to do with bugs at all; you're conflating the issue with irrelevancies.
This is nothing more than an additional evesdroppable side-channel.
Full stop.
As such it in itself provides zero extra security, assuming Kerchoff's principle, and does so _inefficiently_. dst's UDP equivalent is equally secure (zero against the typical theoretical opponent), but far more efficient resource-wise.
Somehow I missed the microsoft address, I don't know how!?!? I'm sure they're all guilty of _something_, yes!
Thanks for the 403 feedback. Initially there were only a few links, but the list just kept growing. As you say, it's not hard to find articles, so I kind of gave up adding new links a while ago, and only every now and then do I add a new one.
What do you mean "they could just make a separate agreement"? It takes _two_ parties to make an agreement. Sveasoft cannot, on their own, make an agreement. The statement is _meaningless_.
In most countries, there's freedom of contract. A _separate_ agreement, which under the GPL you must not be obliged to agree to in order to get access to source, can say almost _anything_. If you agree, in a separate agreement, to pay them money if you exercise your GPL rights, then if you do exercise your GPL rights, then you must pay them. In no way does that affect your rights under the GPL, as it was a _separate agreement_, that you weren't obliged to agree to. They can't, under the GPL, oblige you to enter into any other contract, but likewise the GPL cannot remove your right of freedom of contract. It all swings on the independence of the two agreements.
"you can revoke for any number of other reasons freely, just not for redistributing the source"
The FSF and the GPL say nothing in that regard. A subscription agreement can have any cancellation clause it likes in it./If/ Sveasoft have included something ultra-nazi like "we reserve the right to cancel your subscription without warning and without reason", then they're bulletproof.
But that's an "if" - anyone know what the terms really are?
Slashdot Mirror
For a computational number-theory project I run (PIES) I was doing some preparatory number-crunching work before handing out jobs to clients.
My code is reasonably-optimised portable C, and I set it running on a 533MHz 21164 (from 1996, IIRC?), and a 2.8GHz P4.
They were running at within 5% of the speed of each other. The P4 architecure, unless you diddle in assembly, is pants. (Yes, I compiled using Intel's compiler.)
And given that this Alpha machine has been loaded >1.0 since time began - it's unbelievably rock solid:
ecmnet@megaspaz:/etc/apache$ uptime
21:26:25 up 507 days, 8:41, 20 users, load average: 1.00, 0.96, 0.92
Which you'll never see on an x86 linux machine.
You're confusing "protecting", with "aggression".
Sorry, I know they're long words; just do your best.
FP.
In theory, being from the kind of technical background that I am, I ought to fawn over every column, but, to be honest, I find his usual statements to be a bit feeble, a gentle puff, with no real gusto. He does pull his punches.
Normally.
However, this one has broken that mould. There were no punches pulled, and he completely nailed his colours to the mast. Good on him.
However, I'd be tempted to say that he's even made himself a target of Microsoft lawyers, as he has made allegations which could be, if false, be taken as libelous (or otherwise defamatory). (Not that I believe they are false.)
Will the posse of lethal attack-lawyers be set on him for it? Or will MS just hope it gets forgotten about as quickly as possible?
FP.
From that PDF:
"""
Who would you vote for 2nd choice: Badnarik 47%
Have you ever heard of Badnarik: No 70%
"""
Great way to prioritise candidates, guys!
The US population probably gets what the majority of it deserves.
It's a shame it inflicts it on others too though.
FP.
"""
Some people spend hours every day triaging bugs. Bugzilla is a workflow miracle in the sense that it allows this volume of bugs to be dealt with effectively.
"""
I think you meant "Mozilla is a freaking disaster area if it allows this volume of bugs to exist in the first place".
I'm a vocifereous pro-OSS advocate. Yet even I admit that Mozilla's a steaming pile, an embarassment to the community as a whole.
FP.
You can also check for running daemons in /var/run/*.pid
FP.
Your link points to this obvious nonsense:
"""
Consider what happens when you do a new/malloc: a) the allocator wanders through some lists looking for a slot of the right size, then returns you a pointer. b) This pointer is pointing to some pretty random place.
With GC, a) the allocator doesn't need to look for memory, it knows where it is, b) the memory it returns is adjacent to the last bit of memory you requested. The wandering around part happens not all the time but only at garbage collection. And then (depending on the GC algorithm) things get moved of course as well.
"""
The guy knows nothing about _either_ efficient dynamic memory allocators, nor garbage collectors, he has no right to pass comment on them.
FP.
"""
You have to love a system that requires downtime as part of uptime. How many Linux users have this problem?
"""
I run linux and have a similar "data overload" problem:
phil$ uptime
17:48:38 up 82 days, 5:20, 11 users, load average: 1.00, 1.00, 1.00
The "data overload"? My uptime's actually 497+82=579 days.
The digits 4, 9 and 7 in that order may ring a bell.
FP.
September 14th, eh?
How about Jun 23rd.
That's June 23rd, _1993_
From Google groups:
>>
FP.
Why couldn't your post be nearer the top? I was nearly giving up on this story due to the unbelievable amount of misinformation and general gibbering!
I couldn't agree more with your points, the one thing that stood out most glaringly in the article was the one you pick on too:
"""
One of the key breakthroughs is an "intermediate representation," a kind of lingua franca that gives the software the flexibility to translate from one platform to another.
"""
So one of the breakthroughs is to use a bytecode in the same way that Java has since day 1. That's so not a breakthrough. The breakthrough is to get it _fast_. The claimed 80% is very impressive, unless that's emulating x86/windows on x86/linux, in which case it's crap.
FP.
fx32!
(alpha)linux tried to adopt it as em86, but that seems a dead project, I could never get it to run.
It was bloody clever. My 533MHz 21164 ran like a 266MHz P2 (which was the same vintage or newer).
However, getting a fully 64-bit processor to emulate a crappy 32-bit one was never a clever idea, in the grand scheme of things.
FP.
"Calculating Pi to a billion digits is not floating point math, "
A lgor ithms/fft.html
I call bullshit.
George Woltmann, for his bignum multiplications in GIMPS, uses Complex (i.e. FP) FFTs.
Yves Gallot, for his bignum multiplications in Proth, uses Complex (i.e. FP) FFTs.
Phil Carmody (i.e. me) for his bignum multiplications in ForEis, uses Complex (i.e. FP) FFTs.
Given that Xavier Gourdon hasn't realeased the source to PiFast, we can't know for sure, but given his comments in section 2.2 at
http://numbers.computation.free.fr/Constants/
wouldn't you be likely to infer that Xavier Gourdon for his bignum multiplications in PiFast, uses Complex (i.e. FP) FFTs?
FP.
The point that sems to have not been made is that when you plug in a 487 "coprocessor", it actually takes over the whole system, and your old 486SX is pretty much redundant. i.e. the 487 is a whole 486DX.
FP.
Totally agreed.
Primegen comes with a 'count' function which does no output, that would entirely isolate the CPU-bound component.
However, chosing to not turn on any optimisation for some of those tests means that these benchmarks were the most "synthetic" I've every seen -- they certainly corresponded to nothing real.
FP.
Mandelbrot is not the father of the fractal.
... the stock market.
It's much more appropriate to give that title to the Frenchman Julia.
And Julia discovered fractals while studying...
This was over a freaking century ago! Wake up Mandelbrot - you're _still_ not breaking new ground.
FP.
ZO has been in the dictionary, even small ones, for decades. It's like a yak.
The more important issue is when WIPQOZN is going to be added to the scrabble dictionary?
FP
I don't know what to believe in that article - it is quite poorly researched:
"
Unfortunately, the pins are rather large to be electrical and mechanical devices. Only a few hundred fit on a package that contains a processor with several million transistors. Bandwidth, thus, is constrained.
"
A few hundred? Explain why IBM manage to get more than five _thousand_ pins on their POWER packages.
FP.
This way, when "the enemy" notice weird TCP connection attempts to your box, they'll know that you're using port-knocking, and decide to hammer every single port in random order just on the offchance.
It's a great way of finding out if your TCP stack has memory leaks.
FP.
No.
Listen to what gst said.
This is nothing to do with bugs at all; you're conflating the issue with irrelevancies.
This is nothing more than an additional evesdroppable side-channel.
Full stop.
As such it in itself provides zero extra security, assuming Kerchoff's principle, and does so _inefficiently_. dst's UDP equivalent is equally secure (zero against the typical theoretical opponent), but far more efficient resource-wise.
FP.
That's because you're ignoring Kerchoff's principle.
Your keystrokes can be transmitted over a non-evesdroppable keyed session. Port-knocking can't.
FP.
Somehow I missed the microsoft address, I don't know how!?!? I'm sure they're all guilty of _something_, yes!
;-)
Thanks for the 403 feedback. Initially there were only a few links, but the list just kept growing. As you say, it's not hard to find articles, so I kind of gave up adding new links a while ago, and only every now and then do I add a new one.
Glad you spotted the bias
Phil
My bet is that the virus coder is a lameass linux wanabee.
Why do I say this? Because he reads his own mail as root@wherever, i.e. hasn't worked out what user accounts are for.
He also contributes to open source projects.
How do I know all this - just look at the list of usernames he won't spam. root is excluded and so is *sf.net and *sourceforge*.
Just a guess.
Of course, doing everything as root is the mentality of a WinDoze luser, of course, so maybe you're right after all...
FP.
What do you mean "they could just make a separate agreement"? It takes _two_ parties to make an agreement. Sveasoft cannot, on their own, make an agreement. The statement is _meaningless_.
In most countries, there's freedom of contract. A _separate_ agreement, which under the GPL you must not be obliged to agree to in order to get access to source, can say almost _anything_. If you agree, in a separate agreement, to pay them money if you exercise your GPL rights, then if you do exercise your GPL rights, then you must pay them. In no way does that affect your rights under the GPL, as it was a _separate agreement_, that you weren't obliged to agree to. They can't, under the GPL, oblige you to enter into any other contract, but likewise the GPL cannot remove your right of freedom of contract. It all swings on the independence of the two agreements.
FP.
How?
Millions of dollars of investment in their infrastructure.
Which unsurprisingly cost millions of dollars.
The GPL does not mandate that you mass-produce your source, nor spend millions of dollars on your infrastructure.
FP.
"you can revoke for any number of other reasons freely, just not for redistributing the source"
/If/ Sveasoft have included
The FSF and the GPL say nothing in that regard.
A subscription agreement can have any cancellation clause it likes in it.
something ultra-nazi like "we reserve the right to cancel your subscription without warning and without reason", then they're bulletproof.
But that's an "if" - anyone know what the terms really are?
FP.