AMD planning 45nm 12-Core 'Istanbul' Processor ?

[Robert Myers]

Robert Myers wrote:


Another attempt at twist?

Are they alive? Or not?


Uh, oh.

What does that mean? How old are *you*? The Federal Reserve has done
things that at one time would have been unthinkable and they are just
the beginning. Were it not for those actions, the financial markets
would have collapsed like a house of cards, as they almost did with
the failure of Long Term Capital Management. I don't know about the
actions of other central banks, but I do know what's happened here.
Sovereign wealth funds have been pouring in capital to vulnerable
investment bankers who otherwise would have failed. It's not in their
interests to see US financial markets collapse, either. Did you get
to the chapter titled "Too Big to Fail" yet?

Guy can't fly a plane so he can't drive a car -- it's the same idiotic
logic.

So. What you *should* have said was, "Financial institutions can't
estimate risk in their main line of business, but they know all about
software." You mean: they're overconfident about their main line of
business, so they buy what you tell them about software. The fact
that you're selling to financial institutions is something you never
should have presented as a credential.

Oh, you "know" software security, medical production, software production,
markets, cold war, Edsger Dijkstra, NASA, and whatever else. Typical of
high school kid. And, unfortunately, like majority of high school kids you
can't read.

I can read and write very well. Since English is probably not your
first language, I won't comment on your skills. My first example was
of a failure of software used to estimate risk, not of the risk to the
software itself. I very well knew that. That's why I put stars
around the reference. The fact that we are talking about software to
assess risk and software that is susceptible to risk is apparently too
confusing for you, as you are imagining that I would confuse them as
you apparently want to. The software that NASA used to assess risk
got wildly optimistic answers. That says there is something wrong
with the methodology to assess risk. I've also given examples of
failures of the software itself (flight control software for deep
space probes) and Standard and Poors giving AAA rating to securities
that probably should have had junk bond status. That's *three*
different kinds of software. I hope you can keep them straight.

LOL!

The huge pile of software that's out there was, largely, built to
exploit the capacities of one-processor systems. One-processor
systems are going the way of the dinosaur. The problems with multi-
processor programming are fundamental and make the problem of software
bugs much, much worse. Something has to give.

ROTFL!

You're simply piece of the work. Go on! Keep invenitng what others have
said.

Current methods do work. It doesn't change the fact that these methods do
evolve. But they evolve to make software production more effective not to
satisfy some guy with high school kid attitude and black-white views.

No one has to satisfy me about anything. As to the methods of
software production evolving, that's laughable. What we see are heaps
and heaps of tools and languages, none of them fundamentally better
than what's been in place for decades. I just don't know what to say
to someone who says, "Current methods do work." Manifestly, they
don't.

Robert.

 

[Robert Myers]

Of course you did. Thats you not me who can't see the obvious difference.

Why did you ever say that the financial institutions you work for know
how to estimate risk? Manifestly, they don't. Why should I accept
that they are wizards on risk in a field that is not their specialty
(software) when they screw up so badly in a field that *is* their
specialty (securities)? The expert on software risk here is *you*,
not the people you work for.

Because it's obvious. They would be nonexistant if they didn't.

Besides I do not work for American financial company, so your nonsense does
not apply even if there would be no difference between markets and
software.

The day may come when Eastern Europe drives financial markets. I
don't think we're their yet. Your arguments become ever more arcane.
Well, they may not know how to estimate *American* risk, but they know
how to estimate *Polish* risk.

Uh, oh.

I can't even imagine what to say to you. You think it's all funny.

Yeah. The finacial software will explode releasing about equivalent of 1kt
TNT of energy due to failed o-ring.

Let's see. Pulling some numbers about the latest debt obligation
screwup. Moody's (not S&P, my error) overrated about $4 billion in
debt. One estimate of loss was about 40%. I'll use $2 million for
the value of a human life. The cost of the screwup was about 2000
lives lost. But, you sputter, no lives were actually lost? Think
again. Maybe not 2000 lives right away, but some pretty quickly, as
people discover that a key asset lost 40% of its value practically
overnight. How many car bombs does it take to kill 2000 people?

So your proposed way of (not) making a software would not help either.

I'll stand by my original position. Less software is, almost by
definition, better software.

Risk of loosing shuttle and astronauts was and still is the "acceptable"
risk for NASA.

You really don't know the history of the shuttle program. Being a
shuttle crew member has proven to be riskier than being a combat pilot
in World War II. Had that been understood at the beginning, the
design would have been canned. As it is, the US will be without
manned space flight capability for several years. Imagine how heads
would have rolled if that outcome had been foreseen.

The risk of major terrorist attack was known for long time
before 9/11. As few soviet atomic suitcases are believed to be missing, the
risk of even greater disaster is still non zero.
But there is also a significant risk of simply getting killed in traffic
accident, and that risk is much greater than becoming a victim of terrorist
attack. But life must go on, we simply must accept that risk. Not accepting
them and hiding somewhere is widely regarded as mental disease.

We must accept risk. That doesn't mean that we must accept what you
propose as reasonable. The actual damage of the World Trade Center
was nothing compared to the emotional toll it took. When you build
things like atomic bombs or jumbo jets or globe-spanning financial
networks with hair-trigger mechanisms built in, you're changing the
rules about the kind of risk that humans live with. To say "it's
always been like this and we have to live this way" just isn't
correct.

Robert.

 

[Robert Myers]

Let's see. Pulling some numbers about the latest debt obligation
screwup. Moody's (not S&P, my error) overrated about $4 billion in
debt. One estimate of loss was about 40%. I'll use $2 million for
the value of a human life. The cost of the screwup was about 2000
lives lost. But, you sputter, no lives were actually lost? Think
again. Maybe not 2000 lives right away, but some pretty quickly, as
people discover that a key asset lost 40% of its value practically
overnight. How many car bombs does it take to kill 2000 people?

Yes, I left out a factor of 0.4. That should be 800 lives, not 2000.
No, I can't do arithmetic.

Robert.

 

[Robert Myers]

On a sunny day (Tue, 27 May 2008 09:36:50 -0700 (PDT)) it happened Robert


I can read what you wrote, so it _does_ work.

It may not be perfect, but it never will be.

There are also many ways to build a house, concrete,
bricks, wood, and many tools to do it.
Thats is an advantage, not a disadvantage.
And the right person will use the right tools for the job.

Before the most recent earthquake in China, you could have gone on a
tour of the stricken region and had a regional official tell you, "See
our building codes and our land use planning work." Similar things
would have been said to anyone who worried about San Francisco in the
early 1900's or Chicago before it burned down.

"See, it works" is about the most foolish kind of risk estimation
there is. When the lesson is hammered home in the harshest possible
way (tsunami, earthquake, hurricane, market collapse, shuttle
disaster, Chernobyl, Three Mile Island), there is no place left for
the wishful thinkers to stand. New Orleans "worked," even though much
of it was below sea level. "See, it works."

http://blogs.trb.com/news/politics/blog/katrina-new-orleans-flooding3-2005-thumb.jpg

See?

Building codes, tsunami warning networks, land use regulations,
nuclear safety controls, market circuit breakers and increased
regulation, and better risk analysis all happen the hardest possible
way: whatever it was, it *didn't* work well enough. NASA, I'm afraid,
is beyond hope.

Robert.

 

[Robert Redelmeier]

I should have told you to take up your argument with the
architect I quoted. He works there. You don't. The quote
is easy enough to find. Go argue with him.

Why? You are the one extending the argument.
You need to be able to defend it with competence.

You'd have a lot of work to do to convince me that there is
such a thing as human nature, never mind that it explains
anything, never mind that it is a potential cause of
thanksgiving or regret.

ROTFL! You believe that people always act rationally, homo
economicus? Do you think SUVs were foist onto stupid people?
No -- they bought them willingly for advantages such as higher
seating position and perceived (not actual) safety. In spite
of well-known disadvantages like parking and fuel.

The only reason that we are not in a colossal world-wide
depression right now is because we've gotten smarter about
fiscal and monetary policy and the tools have been created
to allow that knowledge to be acted upon.

Hubris! The only reason we are not in a needed depression
is gross market manipulation by the US FRB and other central
banks who have been serial bubble-blowers: first tech, then
finance, then real-estate and now commodites. Each one bigger
than the last to cushion the fall but worsen the final pop.

I think the x86 monoculture we have is a lamentable state
of affairs.

I am far less convinced we have that, nor that it is on balance
negative. Would a Moto68k or ALPHA monoculture be better?

Am I certain that patients have died in the modern era because
of inadequate oversight of drug companies? Yes, I am.

So am I, but that was not the question. Of course people
have died. The question is whether more people have died
than would have been saved. Many people with HIV are alive
because FDA procedures were waived. Some have died.

Because I acknowledge that regulation of markets is often
necessary does not make me an east coast elitist, never
mind a socialist, never mind a communist.

Strawman! I think you are losing the thread. Where have I
ever accused you of any of these things you seem to consider
objectionable??

[ISO 9000] How burdensome it can be for a small company.

Only if that company was taking shortcuts in the first place.

Capitalism works as long as markets and economies are growing,
but so do Ponzi schemes. I'm not sure I can see the difference.

I take you at your word. You might try to learn about zero-sum
games versus net positive sum games.

I can't begin to imagine how a line-by-line review of software would
accomplish anything. You'll find mistakes, I'm sure. You won't
find all the mistakes and the code base is a moving target.

The code has legs? It writes itself? Have you never heard of
a freeze? Line-by-line is tedious but remarkably effective
at finding security holes. With your vast experience, how
would you suggest quality code be written?

That's true, although the RIAA seems to be proving that
even massive civil disobedience can be dealt with.

You do live in fear. The RIAA is fighting a desparate,
failing rear-guard action. Yes, they hurt a few people
and really want to scare more (like you).


-- Robert

 

[Robert Redelmeier]

Yes, I left out a factor of 0.4. That should be 800 lives, not 2000.
No, I can't do arithmetic.

That is the _least_ of your fallacies. So many in
so few words, are you going for a record?


1) Where does the $2 million come from? I know of no generally
accepted value of human life and the subject is highly controversial.
Your $2 million seems very low, but I guess you'd play Russian
Roulette to win a house.

2) Paperhanging is a zero net-sum game (unlike primary
investing). Somebody made the $1.6 billion that others lost.
Developers, bankers, builders, tradesmen, service workers.

3) The loss (or gain) was set when the paper was written,
not when it was sold with negligent ratings. Repayment by
known marginal borrowers was obviously a bet on continually
rising real-estate prices.

4) It is far from certain who (other than taxpayers) will actually
suffer the losses. Much of the paper is held by guaranteed pension
funds. Eurobanks squawked first because they actually worry, and
occasionally have auditors and regulators who will call them out.

5) Moody's disingenuously blame a "computer glitsch" when their
methodology and algorithms were negligent from the start which
no amount of software quality verification would catch:
http://www.tilsonfunds.com/Einhorn-4-08.pdf

6) The real effect of this and other debacles is to make people
more cautious about markets. Pay more attention to risk and
price it accordingly. IMHO, a sorely needed development.



-- Robert

 

[Sebastian Kaliszewski]

What does that mean?

Don't speak about things you know little about.

Others have answered your nonsense already. Then add to that that its your
central bank and your government which worked hard not so long ago to make
the situation as it is now (for short term gains). They have screwed so
they have to intervene now.

So. What you *should* have said was, "Financial institutions can't
estimate risk in their main line of business, but they know all about
software."

You're piece of the work. They *can* & *do* estimate risk in their main line
of business. Otherwise they would be dead long time ago.

I can read and write very well.

It's apparently disputable.

Since English is probably not your
first language, I won't comment on your skills.

I never claimed I could write well. But I a least try to stick to the
subject.

My first example was
of a failure of software used to estimate risk, not of the risk to the
software itself.

Which is completely off side.

I very well knew that.

So you intentionally tried to twist..

That's why I put stars
around the reference. The fact that we are talking about software to
assess risk and software that is susceptible to risk is apparently too
confusing for you,

Nope. But these things have little in common.

as you are imagining that I would confuse them as
you apparently want to.
Whatever.

The software that NASA used to assess risk
got wildly optimistic answers.

Whatever. What all of that has to software production methodology?

That says there is something wrong
with the methodology to assess risk.

Risk of what?

I've also given examples of
failures of the software itself (flight control software for deep
space probes) and Standard and Poors giving AAA rating to securities
that probably should have had junk bond status. That's *three*
different kinds of software. I hope you can keep them straight.

The huge pile of software that's out there was, largely, built to
exploit the capacities of one-processor systems. One-processor
systems are going the way of the dinosaur.

Individual cores of multicore machines are not slower than earlier single
cores. They are much faster in fact. The huge pile of software out there
is "happy" with fraction of performance of that single core. So the soft
utilises less than 25% of power of that Core2 Quadro? Who cares.

The problems with multi-
processor programming are fundamental and make the problem of software
bugs much, much worse. Something has to give.

Whatever. It so happens that software I typically work on is designed to
work in thousands concurrent contexts. 100 core machines? Nice, our clients
will be able to reduce hardware costs and run our software on few machines
instead of few hundreds.

Yes, classical multithreaded programming is most likely to give. It scales
poorly and introduces tons of heisenbugs. But there are already known ways
to deal with that. Those who made only single-context programs will have to
learn new things -- but it was allways that way in software development --
you learn new things all the time or you're off.

No one has to satisfy me about anything. As to the methods of
software production evolving, that's laughable.

It's laughable only for the clueless.

What we see are heaps
and heaps of tools and languages, none of them fundamentally better
than what's been in place for decades.

Fundamentally is not required. Simply better is enough. Black-whites with
lack of grip in reality are not satisfied -- but reality does not care.

I just don't know what to say
to someone who says, "Current methods do work." Manifestly, they
don't.

ROTFL!!!

So how you managed to even post that nonsense to this newsgroup. If the
metods to produce software to post didn't work there should be no
software...


Sebastian Kaliszewski

 

[Sebastian Kaliszewski]

Jan Panteltje wrote:
[...]
You're trying to show reality to a man who can't grip it. It's an excercise
in futility, I'm affraid. I'm slowly giving up hope.

rgds
Sebastian Kaliszewski

 

[Sebastian Kaliszewski]

I think the x86 monoculture we have is a
lamentable state of affairs.

Oh. That sums it up. You care of nerd delusions not for reality.

In reality it simply does not matter. X86 or not is a little technical
detail which is of interest to a few technical people, but which simply
does not matter for the customer. What matters for her/him is the ability
to run the software they have licences for.


If someone laments about such things like x86 I could say one thing: get a
life!


Sebastian Kaliszewski

 

[Sebastian Kaliszewski]

Why did you ever say that the financial institutions you work for know
how to estimate risk? Manifestly, they don't.

Manifestly they do.

Why should I accept
that they are wizards on risk in a field that is not their specialty
(software) when they screw up so badly in a field that *is* their
specialty (securities)? The expert on software risk here is *you*,
not the people you work for.

They are experts of business risk, including risk a software has for their
business weighted against not having that software, and having similar
functionality but for 10 times the price and having to wait for much longer
for it. Especially when your competition wants to do the same, time to
market is a key factor. And that's the reality.

The day may come when Eastern Europe drives financial markets. I
don't think we're their yet. Your arguments become ever more arcane.
Well, they may not know how to estimate *American* risk, but they know
how to estimate *Polish* risk.

Well, the institution is not Polish. It prospers. So talk about European
risk.

I can't even imagine what to say to you. You think it's all funny.

The funny stuff is our performance here.

Let's see. Pulling some numbers about the latest debt obligation
screwup. Moody's (not S&P, my error) overrated about $4 billion in
debt. One estimate of loss was about 40%. I'll use $2 million for
the value of a human life. The cost of the screwup was about 2000
lives lost. But, you sputter, no lives were actually lost? Think
again. Maybe not 2000 lives right away, but some pretty quickly, as
people discover that a key asset lost 40% of its value practically
overnight. How many car bombs does it take to kill 2000 people?

Another Robert has already answered that better than I could.

I'll stand by my original position. Less software is, almost by
definition, better software.

And no software means unbelivably good software, as elements of empty set
have any property you want.

You have no grip in reality. In reality what matters is good enough software
for low enough price. But you don't get it. And I'm loosing hope you ever
will.

You really don't know the history of the shuttle program. Being a
shuttle crew member has proven to be riskier than being a combat pilot
in World War II.

And you pulled that number from what?

Anyways, current life expectancy of combat airplane mission in case of
more-or-less equal powers at war is 1.5, while its 50 for shuttle. That is
a big difference.

Had that been understood at the beginning, the
design would have been canned.

They would go for a better & safer design with lesser capacity (as
originally planned, but military wanted high capacity space truck).

As it is, the US will be without
manned space flight capability for several years.

Not for the first time. Besides they do acklowledge that risk is about 2%
(as historical records show) and still fly.

Imagine how heads
would have rolled if that outcome had been foreseen.

Shuttle is the most trustworthy human space transportation system despite
two fatal incidents. The comparable one is russian Souyz, but it also had
it's death toll.


[...]
Sebastian Kaliszewski

 

[Robert Myers]

Manifestly they do.


They are experts of business risk, including risk a software has for their
business weighted against not having that software, and having similar
functionality but for 10 times the price and having to wait for much longer
for it. Especially when your competition wants to do the same, time to
market is a key factor. And that's the reality.



Well, the institution is not Polish. It prospers. So talk about European
risk.

My point was that you now have to argue that the risk you are facing
is somehow different from what I know of what has happened on Wall St.

And no software means unbelivably good software, as elements of empty set
have any property you want.

You have no grip in reality. In reality what matters is good enough software
for low enough price. But you don't get it. And I'm loosing hope you ever
will.

There is *zero* chance that I will see the world the way you do, so if
you're trying to get that outcome, you're on a fool's errand. Saying
that the less software the better is a kind of rhetorical device.
There are plenty of old-timers who would agree with me. You're spot
on with the time-to-market argument, and that, not any realistic
assessment of risk, is what drives software development right now.

And you pulled that number from what?

Public discussion right after Columbia. In the case of Challenger,
NASA just flat out lied to itself. In the case of Columbia, NASA did
what I had seen very early in the program, which was to draw
conclusions based on inadequate or non-existent analysis. Neither
possibility had ever been in anyone's estimations of risk, never mind
that the chance of it happening was estimated correctly.

Anyways, current life expectancy of combat airplane mission in case of
more-or-less equal powers at war is 1.5, while its 50 for shuttle. That is
a big difference.

So what has more or less equal powers got to do with it? The US and
the UK, not to their combined credit, flew many, many missions against
poorly defended cities in WWII. I can't, off-hand, think of any US
engagement where a pilot was, on average, dead after 1.5 missions.

They would go for a better & safer design with lesser capacity (as
originally planned, but military wanted high capacity space truck).

The point being that the risk would not have been acceptable.

Not for the first time. Besides they do acklowledge that risk is about 2%
(as historical records show) and still fly.

You don't really understand the centrality of space operations to the
US military posture. For a while, the US couldn't get anything into
orbit because all the chips had been put on that stupid shuttle. And
the key to the entire enterprise was faulty estimation of risk.

Shuttle is the most trustworthy human space transportation system despite
two fatal incidents. The comparable one is russian Souyz, but it also had
it's death toll.

The piece you're missing here is that the shuttle program could never
have been sold as it was. It was supposed to be a vehicle for
routinely delivering satellites on orbit. Those satellites, many of
which are never publicly discussed, are crucial to both US and
European security. There was a time that the US was so paranoid that
I couldn't even have said that, even though it was really no big
secret.

Robert.

 

[Robert Myers]

That is the _least_ of your fallacies. So many in
so few words, are you going for a record?

1) Where does the $2 million come from? I know of no generally
accepted value of human life and the subject is highly controversial.
Your $2 million seems very low, but I guess you'd play Russian
Roulette to win a house.

Where does *that* snarky comment come from? I pulled the first number
I could find, because it doesn't matter if it's off by a factor of
five. The difference between an entire human life and part of a human
life (which is what you take when you kill adults) is about a factor
of 2.

2) Paperhanging is a zero net-sum game (unlike primary
investing). Somebody made the $1.6 billion that others lost.
Developers, bankers, builders, tradesmen, service workers.

You're seeing a distinction that I don't. Criminals get to spend
their money, too.

3) The loss (or gain) was set when the paper was written,
not when it was sold with negligent ratings. Repayment by
known marginal borrowers was obviously a bet on continually
rising real-estate prices.

If you wave your arms any faster, you're going to take off. For one
thing, the actual ability of borrowers to repay isn't the biggest,
baddest issue. The biggest, baddest issue is that, at any given
instant of time, the paper is worth no more or less than what people
think it's worth. The mistake at Moody's compounded a central
problem: no one could evaluate each other's assets reliably, so
lending based on those assets as collateral stopped. The house of
cards collapsed, and it could have collapsed completely independently
of any underlying economic activity (whether borrowers actually repay
or not).

These problems become apparent when someone is spooked. The problem
isn't that investors are spooked. That's always going to happen. The
problem is that people are making fabulous livings by gutting the
structure of markets. When investors do get spooked, the real cost is
either monetized or paid by the taxpayers.

4) It is far from certain who (other than taxpayers) will actually
suffer the losses. Much of the paper is held by guaranteed pension
funds. Eurobanks squawked first because they actually worry, and
occasionally have auditors and regulators who will call them out.

You can monetize anything. Been to the store recently?

5) Moody's disingenuously blame a "computer glitsch" when their
methodology and algorithms were negligent from the start which
no amount of software quality verification would catch:
http://www.tilsonfunds.com/Einhorn-4-08.pdf

A quick scan of the document you offer gives nothing like the
specifics you claim. There are a number of problems with risk
estimation. Software quality is only one and maybe not the biggest
one, but I haven't seen any claim that the Moody's problem was
anything other than software.

6) The real effect of this and other debacles is to make people
more cautious about markets. Pay more attention to risk and
price it accordingly. IMHO, a sorely needed development.

Isn't that what I've been talking about?

Robert.

 

[Robert Myers]

Don't speak about things you know little about.

Others have answered your nonsense already.

The more you say things like that, the less credible you become.

Then add to that that its your
central bank and your government which worked hard not so long ago to make
the situation as it is now (for short term gains). They have screwed so
they have to intervene now.

I don't claim to understand what's happening outside the US, except
that the debt crisis is worldwide. Jerome Kerviel was working for a
French institution. As far as I could tell, he was playing the same
fool's game: trying to make big money by assuming arbitrarily large
risk. We live in worldwide markets. "They" is increasingly "us."

You're piece of the work. They *can* & *do* estimate risk in their main line
of business. Otherwise they would be dead long time ago.

Well, no. They've discovered an elaborate scheme for transferring
risk around, taking a profit as it changes hands.

Heard of Bear, Stearns? Once one of the most feared and ruthless
investment banks on Wall Street:

http://online.wsj.com/article/SB121193290927324603.html (subscription,
but since you know so much about the business, I'm sure you have one).

Here's a New York Times link:

http://dealbook.blogs.nytimes.com/2008/05/28/for-bear-fear-in-the-market-was-fatal/?hp

You have to register, but it's free.

Which is completely off side.

You made broad claims about risk estimation. I've made broad counter-
claims. Neither NASA nor Wall Street knows as much as it thinks it
does. There are all kinds of problems. So your sweeping claim that
everything is okay because you work for people who know how to
estimate risk is laughable for a number of reasons:

1. The methodology of risk estimation is not in good shape.
Historically, people have made claims that haven't been supported by
experience.

2. In particular, people who have "critical" applications have made
big screwups: NASA and investment bankers--and Arianspace.

3. In particular, the kinds of people you claim for credibility are
not credible. You'll search the pages of the Wall Street Journal for
a long time without finding such an admission because the claimed
credibility is the source of so many paychecks, including yours.
Because lots of people are doing something a particular way doesn't
make it right.

4. Actual and very serious problems have occurred.

So you intentionally tried to twist..

There was no intent to deceive you.

You keep calling me derogatory names, but maybe you want to step back,
calm down, and listen.

The big, underlying event here is computers. Microprocessors were a
game changer, but not as big as computers themselves. Computers made
possible calculations that were previously unthinkable. They
introduced the possibility of strategies so complicated that no
ordinary mortal could understand them or keep up with them (e.g.
program trading).

The ability to do those calculations and to implement those strategies
has introduced a new kind of risk, which is, ironically, the risk that
your method of estimating risk is faulty. Engineers have always
understood that elimination of risk is impossible and that estimation
of risk often just guesswork. What's different about financial
markets is that people are becoming more aggressive with risk because
computers give them the illusion that they understand it better than
they do.

*That's* why I have talked not only about the process at risk (writing
software), but about the software used to estimate risk. Probably it
would have been less confusing if I had talked about the methodology
used to estimate risk, but the fact that it's often being done by
computers is telling: we're removing ourselves ever further from the
actual problem.

Whatever. What all of that has to software production methodology?

Risk estimation is risk estimation. Or are you telling me that
financial software has invented its own methods for estimating risk?
That's reassuring.

Risk of what?

So, let's see. There's a theory of risk for each human activity?
There's something special about software development that makes it
more analyzable than other human activities?

Individual cores of multicore machines are not slower than earlier single
cores. They are much faster in fact. The huge pile of software out there
is "happy" with fraction of performance of that single core. So the soft
utilises less than 25% of power of that Core2 Quadro? Who cares.

Plenty of people will care. It's the same as the time-to-market
argument. If your competitor can do it better or faster, you're
screwed.

Whatever. It so happens that software I typically work on is designed to
work in thousands concurrent contexts. 100 core machines? Nice, our clients
will be able to reduce hardware costs and run our software on few machines
instead of few hundreds.

Yes, classical multithreaded programming is most likely to give. It scales
poorly and introduces tons of heisenbugs. But there are already known ways
to deal with that. Those who made only single-context programs will have to
learn new things -- but it was allways that way in software development --
you learn new things all the time or you're off.

Only time will tell. You see a gradually changing landscape. I see
an earthquake.

It's laughable only for the clueless.

There is no way that I can see for currently existing software methods
to "evolve." They're built around languages that have no built-in
semantics for concurrency.

Fundamentally is not required. Simply better is enough. Black-whites with
lack of grip in reality are not satisfied -- but reality does not care.

I'm not convinced that many of the tools are even better. Simply
labeling thinking as black and white is lazy. Yes, I have a
mathematician's view of some things. In fact, I take a
mathematician's point of view whenever I can. In mathematics, things
are usually either right or wrong, although, ironically, mathematics
has started to have the same class of problems. I'm not sure why
"gray" thinking is being advanced as the appropriate mindset for
working with computers. If you're so stuck in your mindset that you
can't understand what I'm saying, you can't compensate by calling me
names and applying labels.

ROTFL!!!

So how you managed to even post that nonsense to this newsgroup. If the
metods to produce software to post didn't work there should be no
software...

"It works most of the time" is not an acceptable standard of quality.

Robert.

 

[Robert Myers]

Oh. That sums it up. You care of nerd delusions not for reality.

In reality it simply does not matter. X86 or not is a little technical
detail which is of interest to a few technical people, but which simply
does not matter for the customer.

It's up to the technical people to do the worrying for people who
don't know better.

Technical monoculture is dangerous. Were I selling Power, Sparc, or
Itanium systems, I'd be hammering away at that. If you have to check
x86 with x86 or to protect x86 with x86, there are built-in problems.
If you can't see that, I'm not going to try to explain it to you.

Robert.

 

[Robert Myers]

On a sunny day (Tue, 27 May 2008 13:21:56 -0700 (PDT)) it happened Robert
<[email protected]>:


I am not sure about the point you want to make,
Sure, if I design an electronic digital instrument, like I have here,
I do _not_ take into account (and this thing is embedded full of software, in asm actually),
that in 2038 or so, a big comet may hit it!

Those kind of calamities cannot be taken into the design idea.

Sure, I agree (been reading this thread now a while) that in case of
for example NASA, the original mars lander, errors were made, NASA knows that,
12 or more design changes were made, Phoenix _works_, they learned, and also
had communications during landing to see what went wrong, if anything, this time,
prepared to learn..

In Europe there is a fault line south of Gibraltar, many hundreds of years ago
there was a big earth quake there that flattened houses up into the north of France.

Does that mean everything is now build to withstand a force 8 earthquake?
No, it is cheaper to rebuild.

You can't bring dead people back to life. California, where I lived
for about a decade, has frequent quakes and evolving building codes
that increasingly mitigate damage. California is still not ready for
the "big one," 8.0 on the Richter scale. Fifty percent chance of that
within the next thirty years according to seismologists (who are
burning up a lot of cycles studying the problem). Saying that it's
cheaper to rebuild would get you laughed out of the room in
California. I can only imagine that you've never been in a tall
building during a big quake or even been woken up by one. How many
dead from such a quake in China? Will we ever know? People may not
be able to afford to emulate California, but anyone at significant
risk pays attention.

There is the economic factor, and also the probability factor.
Software, and also financial software (I wrote some), is designed to do a specific job.
_How_ you use the tool, and the risks you want to take, is up to you.
Much of the financial world is simply gambling anyways,
The chances of winning are about 50%, bit more if you are good, a lot
better then 1 in 100 million for a big lotterie, so they take the risks,

In the case of US financial institutions, they are playing with other
people's monopoly money. I doubt if Europe is different. We have an
economic system that incentivizes recklessness. Am I supposed to
admire it?

Yea, part of the Netherlands once was flooded too, they build dikes,
and an elaborate protection system.
Still some comet plunging in the North Sea would wash away things.
Accepted risk.

Not everyone sees it that way. The energy going into tracking rogue
objects that might hit the earth has increased dramatically, and plans
for dealing with the supposedly unthinkable have been floated.

NASA is absolutely great, it learned from its mistakes, what more can you want.

NASA has been learning from its mistakes, over and over and over again
for decades. To be fair about it, the real problem may be in
Washington with OMB and the Congress. The US is no longer really
prepared to fund its own ambitions.

Robert.

 

[Robert Myers]

Why? You are the one extending the argument.
You need to be able to defend it with competence.

I don't think you took my meaning. That was a take it or leave it
proposition. The cited someone in the industry whom I take to be
knowledgeable. I wasn't interested in analyzing the remark before,
and
I'm not interested now.

ROTFL! You believe that people always act rationally, homo
economicus? Do you think SUVs were foist onto stupid people?
No -- they bought them willingly for advantages such as higher
seating position and perceived (not actual) safety. In spite
of well-known disadvantages like parking and fuel.

Who ever said anything about rationally? You can bet that I never
have. "Human nature," whatever you take it to be, is just as
illusory
as the "rational actors" of economics. The human brain can do very
strange things, singly, in groups, and especially in Usenet threads.

Hubris! The only reason we are not in a needed depression
is gross market manipulation by the US FRB and other central
banks who have been serial bubble-blowers: first tech, then
finance, then real-estate and now commodites. Each one bigger
than the last to cushion the fall but worsen the final pop.

I think you are agreeing with me that our economy is a Ponzi scheme.

I am far less convinced we have that, nor that it is on balance
negative. Would a Moto68k or ALPHA monoculture be better?

I don't think a monoculture of any kind is a good idea.

So am I, but that was not the question. Of course people
have died. The question is whether more people have died
than would have been saved. Many people with HIV are alive
because FDA procedures were waived. Some have died.

HIV is a special case. For one thing, it demonstrates how active
consumer involvement can make a difference. It's not particularly a
gold star for the FDA.

[ISO 9000] How burdensome it can be for a small company.

Only if that company was taking shortcuts in the first place.

That's so funny. Slashdot just had an article about TJ Maxx firing
an
employee for revealing that the company had done nothing to change
its
security practices after one of the most massive data breaches ever.
What company doesn't take short cuts, legally or otherwise? Aviation
accidents that involve maintenance issues routinely turn up falsified
maintenance records.

I take you at your word. You might try to learn about zero-sum
games versus net positive sum games.

That's exactly how a Ponzi scheme works: as money is pouring in, it's
a
positive sum game. People are actually making money. It's stops
working when the pyramid stops growing, and that's exactly what
happens
with capitalism. Everyone can get richer while non-industrialized
economies industrialize, but then the inevitable happens: there's no
inexhaustible supply of resources to exploit, so countries like China
just become another competitor for scarce resources. Someone on the
radio just said that China can buy the food off half the tables in
the
world. China wants coal and hydrocarbons, and it wants to burn
them.
If it's a limited-sum game, you eventually hit that ceiling, no
matter
what.

The only potential out is technology, but all the smart people are
going into finance so that they can run your kind of positive sum
game.
I assume that isn't what you meant to propose, but that seems to be
where your logic leads.

would

The code has legs? It writes itself? Have you never heard of
a freeze? Line-by-line is tedious but remarkably effective
at finding security holes. With your vast experience, how
would you suggest quality code be written?

http://en.wikipedia.org/wiki/Leslie_Lamport

http://www.usingcsp.com/



You do live in fear. The RIAA is fighting a desparate,
failing rear-guard action. Yes, they hurt a few people
and really want to scare more (like you).

They don't scare me because the only thing I use bittorrent for is
free software.

Robert.

 

[Sebastian Kaliszewski]

It's up to the technical people to do the worrying for people who
don't know better.

Technical monoculture is dangerous. Were I selling Power, Sparc, or
Itanium systems, I'd be hammering away at that.

Oh, using set of agreed on road sign icons is dangerous! LOL!

If you have to check
x86 with x86 or to protect x86 with x86, there are built-in problems.

Nonsense. x86 is just a set of instructions. It does not matter unless they
could no be implemented effectively (they can & are, of course) or they
would disallow translating higher level languages to them (x86 is the
instruction set with widest complier & interpreter support).

If you can't see that, I'm not going to try to explain it to you.

Good. I don't want your delusions explained. I'm not your psychoanalyst.


Sebastian Kaliszewski

 

[Sebastian Kaliszewski]

You can't bring dead people back to life. California, where I lived
for about a decade, has frequent quakes and evolving building codes
that increasingly mitigate damage. California is still not ready for
the "big one," 8.0 on the Richter scale. Fifty percent chance of that
within the next thirty years according to seismologists (who are
burning up a lot of cycles studying the problem). Saying that it's
cheaper to rebuild would get you laughed out of the room in
California. I can only imagine that you've never been in a tall
building during a big quake or even been woken up by one. How many
dead from such a quake in China? Will we ever know? People may not
be able to afford to emulate California, but anyone at significant
risk pays attention.

But for many of those people what they have was the only option. They simply
could not afford 8 Richter quake resistant buildings (And there are 9
Richter quakes as well). So what is your proposition for them? Live on the
trees? Or perofrom mass suicide and ease population growth problems of
their country?

Not everyone sees it that way. The energy going into tracking rogue
objects that might hit the earth has increased dramatically, and plans
for dealing with the supposedly unthinkable have been floated.

Floated plans is a bit too little. And we watch to little as well. But given
10 years time from observation to actual disaster were probably able to
tackle the problem. But there are Earth based disater scenarios as well.
Yet millions of people live close to active (or recently active) vulcanoes,
under the sea level, etc...

Sebastian Kaliszewski

 

[Robert Redelmeier]

Where does *that* snarky comment come from?

My apologies, I ought to have known you were a bit slow. I must
be too, for the good Sebastian has been pointing you out repeatedly.

A $2 million value on human life is numerically equivalent to
playing standard Russian Roulette in a revolver with one bullet
that will blow your brains out and 5 empty chambers that would
win you a $400k house. Since you deny being "an east coast
elitist" and you use this value for others, I presume you apply
it to your own life as well. I take you at your word.

You're seeing a distinction that I don't. Criminals get
to spend their money, too.

Not everyone involved was criminal. Who do you think was?

If you wave your arms any faster, you're going to take off.

If that impresses you, you must be apoplectic when people
really go out on a limb. Wait ...

For one thing, the actual ability of borrowers to repay isn't
the biggest, baddest issue. The biggest, baddest issue is
that, at any given instant of time, the paper is worth no
more or less than what people think it's worth. The mistake
at Moody's compounded a central problem: no one could evaluate
each other's assets reliably, so lending based on those assets
as collateral stopped. The house of cards collapsed, and it
could have collapsed completely independently of any underlying
economic activity (whether borrowers actually repay or not).

.... uhm, err ... and you were complaining of handwaving???
This is an ineluctible example. The old "perception versus
reality" false dichotomy.

A quick scan of the document you offer gives nothing
like the specifics you claim.

You scan too quickly -- try page 3: "He [recently retired
senior executive of a large rating agency] told me the group
covering the investment banks is 3 or 4 people, and they have
to cover all of the banks. ... they don't even try to look at
the actual portfolio. ... [they] don't have a [risk] model."


-- Robert

 

[Sebastian Kaliszewski]

The more you say things like that, the less credible you become.

I won't repeat what others have rightly said, as it doesn't bring anything
new.

I don't claim to understand what's happening outside the US, except
that the debt crisis is worldwide. Jerome Kerviel was working for a
French institution. As far as I could tell, he was playing the same
fool's game: trying to make big money by assuming arbitrarily large
risk. We live in worldwide markets. "They" is increasingly "us."

You talk here about internal management problem of the institution. The guy
has broken the rules, exploiting poor internal controll. His risk was
similar to a risk taken by a thief. IOW he abused his company. Such stuff
does happen.

Well, no. They've discovered an elaborate scheme for transferring
risk around, taking a profit as it changes hands.

Heard of Bear, Stearns? Once one of the most feared and ruthless
investment banks on Wall Street:

The risk of getting out of business is accepted risk for any business to
begin with. For examply in my local part of the world risk is a part of
legal definition (which has tax consequences -- making business is taxed
differently that working for someone other).

Besides, management is well paid till the collapse. So their risk is ...
loss of job (typically temporary) and maybe reputation.

You made broad claims about risk estimation. I've made broad counter-
claims. Neither NASA nor Wall Street knows as much as it thinks it
does. There are all kinds of problems. So your sweeping claim that
everything is okay because you work for people who know how to
estimate risk is laughable for a number of reasons:

1. The methodology of risk estimation is not in good shape.
Historically, people have made claims that haven't been supported by
experience.

But here they have support by experience as well. If the design methodology
is right then most errors have well understood costs. Those which are not
well understood have probability lower than many more important events
which would be even more deadly anyway.

Software erasing all the data due to error (not due to malice) could
theoretically happen, but it's much more probable that admin erroneously
does rm -rf in wrong directory ("Oh, I just wanted to delete old logs...").

2. In particular, people who have "critical" applications have made
big screwups: NASA and investment bankers--and Arianspace.

Arianespace soft was made according to all the stuff you want to be employed
to make all the soft. It failed. It failed because of changed external
conditions (long after is was made).

3. In particular, the kinds of people you claim for credibility are
not credible. You'll search the pages of the Wall Street Journal for
a long time without finding such an admission because the claimed
credibility is the source of so many paychecks, including yours.
Because lots of people are doing something a particular way doesn't
make it right.

Whatever. They run that buiness for long time, and this business makes
money. And it makes money because they rightly estimate loads of different
risks. Stuff like isnurance companies is based on estimating risk. It's
their job.

4. Actual and very serious problems have occurred.

So what? There is allways a risk of actual and very serious problems.

There was no intent to deceive you.

You keep calling me derogatory names, but maybe you want to step back,
calm down, and listen.

The big, underlying event here is computers. Microprocessors were a
game changer, but not as big as computers themselves. Computers made
possible calculations that were previously unthinkable. They
introduced the possibility of strategies so complicated that no
ordinary mortal could understand them or keep up with them (e.g.
program trading).

The ability to do those calculations and to implement those strategies
has introduced a new kind of risk, which is, ironically, the risk that
your method of estimating risk is faulty.

The main risk is that simply model (or metamodel) is wrong. Stuff like
technical analysis based trading is known for long time. It was done by
hand. It fails quite frequently? So be it. More complex models work better,
are verified on past data, etc. But there is no proof they are correct, as
correctess is impossible when dealing with Schroedinger Cat on steroids.

That the risk estimation could be faulty is a distant second (or even not
second) here.

Engineers have always
understood that elimination of risk is impossible and that estimation
of risk often just guesswork. What's different about financial
markets is that people are becoming more aggressive with risk because
computers give them the illusion that they understand it better than
they do.

People are becoming more aggresive with risk due to apropriately big gains
possible. You don't want to appreciate things like human nature, but stuff
like lotteries does work despite it's on average a loosing proposition to
everybody but the lottery holder. Yet people play, counting for very, very
small chance of winning. In business simply going out of business is
accepted risk. And allways has been.

*That's* why I have talked not only about the process at risk (writing
software), but about the software used to estimate risk. Probably it
would have been less confusing if I had talked about the methodology
used to estimate risk, but the fact that it's often being done by
computers

Its not done by computers but by use of computers. Models (or at least
metamodels) are still human creation.

is telling: we're removing ourselves ever further from the
actual problem.
Risk estimation is risk estimation. Or are you telling me that
financial software has invented its own methods for estimating risk?
That's reassuring.


So, let's see. There's a theory of risk for each human activity?
There's something special about software development that makes it
more analyzable than other human activities?

That software is a closed system with limited & controlled interface to the
world. Software products are probably the most complex constructs humans
ever made, but it was possible to make them because software has many
unique features. It is not material so no material problems apply directly
to it (and indirect effects can be and are abstracted away into relatively
small number of classes of problems), it has no material cost, etc. It's
environment is under full controll.

Plenty of people will care. It's the same as the time-to-market
argument. If your competitor can do it better or faster, you're
screwed.

Older software is already made and sold. So all we should care about is new
software. And only if execution speed of that software is important at
all.If the softwarehas to wait for disk or network then until single core
is not capable of dealing with disk or network fast enough, program could
stay on one core.

Only time will tell. You see a gradually changing landscape. I see
an earthquake.

The enormous mass of existing software makes earthquake impossible. More and
more programs which would benefit from multiple cores will be made to make
use of that. Those who needed such stuff 10, 20 and more years ago already
paved the road. So people will realize that C with system threading is a
poor tool for that? Good, it's high time to realize that. Stuff like
Erlang, CSP and Occam (based on it) is not new.

There is no way that I can see for currently existing software methods
to "evolve." They're built around languages that have no built-in
semantics for concurrency.

Nope. Even Java has (poor as they are, but they are there) semantics for
concurency. Erlang was already used industrially many years ago (and that
one has concurrency in its heart, allmost everything is concurrent in it).
Stuff like MPI is widely used in HPC allowing non-cocnurrent languages to
do concurrent work. Etc, etc, etc.

I'm not convinced that many of the tools are even better.

It's enough that there are some better tools.

"It works most of the time" is not an acceptable standard of quality.

There is no other standard possible. The point is that "most" should be high
enough. And that level differs according to application. And higher level
costs more.

Sebastian Kaliszewski