"cell" architecture is all about local memory
on
IBM Releases Cell SDK
·
· Score: 4, Informative
The "cell" processors have fast access to local, unshared memory, and slow access to global memory. That's the defining property of the architecture. You have to design your "cell" program around that limitation. Most memory usage must be in local memory. Local memory is fast, but not large, perhaps as little as 128KB per processor.
The cell processors can do DMA to and from main memory while computing.
As IBM puts it,
"The most productive SPE memory-access model appears to be the one in which a list (such as a scatter-gather list) of DMA transfers is constructed in an SPE's local store so that the SPE's DMA controller can process the list asynchronously while the SPE operates on previously transferred data." So the cell processors basically have to be used as pipeline elements in a messaging system.
That's a tough design constraint. It's fine for low-interaction problems like cryptanalysis. It's OK for signal processing. It may or may not be good for rendering; the cell processors don't have enough memory to store a whole frame, or even a big chunk of one.
This is actually an old supercomputer design trick. In the supercomputer world, it was not too successful; look up the the nCube and the BBN Butterfly, all of which were a bunch of non-shared-memory machines tied to a control CPU. But the problem was that those machines were intended for heavy number-crunching on big problems, and those problems didn't break up well.
The closest machine architecturally to the "cell" processor is the Sony PS2. The PS2 is basically a rather slow general purpose CPU and two fast vector units. Initial programmer reaction to the PS2 was quite negative, and early games weren't very good. It took about two years before people figured out how to program the beast effectively. It was worth it because there were enough PS2s in the world to justify the programming headaches.
The small memory per cell processor is going to a big hassle for rendering. GPUs today let the pixel processors get at the frame buffer, dealing with the latency problem by having lots of pixel processors. The PS2 has a GS unit which owns the frame buffer and does the per-pixel updates. It looks like the cell architecture must do all frame buffer operations in the main CPU, which will bottleneck the graphics pipeline. For the "cell" scheme to succeed in graphics, there's going to have to be some kind of pixel-level GPU bolted on somewhere.
It's not really clear what the "cell" processors are for. They're fine for audio processing, but seem to be overkill for that alone.
The memory limitations make them underpowered for rendering.
And they're a pain to program for more general applications.
Multicore shared-memory multiprocessors with good cacheing look like a better bet.
This is just "Application service providers again. That's a lousy business, with many players, none of whom make much money. Businesses hate those things. They cost by the month, and somebody else can hold your data hostage.
Microsoft may be able to succeed in this by rigging their OS so that it only works right with their ASP systems. But that didn't work for MSN.
Nor was the Microsoft Connected Services Framework a success.
Now if we can just get Google to move blogs downward in searches, that will be a big help. I'm tired of having to track through three layers of bloggers to get to the actual info. Slashdot stories have been suffering from this problem lately. (Earth to Slashdot: we need the ability to moderate stories.)
Yes, sometimes news breaks on a blog, but mostly, it's just drivel.
Motorola put out a clunky phone with iTunes to test the market, get something out the door, and get a field test. The real product is just coming out - the Motorola RAZR V3i with iTunes capability..
Now that Motorola has the hardware working, they can consider cutting Apple out of the loop. By, say, cutting a deal with WalMart.
The University of California at Berkeley won't accept for credit high school biology courses that teach intelligent design. If you want to get into the life sciences or medicine, get out of Kansas schools.
Drivers outside the kernel should be fully supported, at least for USB, FireWire, and printer devices. There's no reason for trusted drivers for any of those devices, since the interaction with memory-mapped and DMA hardware is at a lower, generic level.
Actually, all drivers should be outside the kernel, as in QNX, and now Minix 3. But it's probably too late to do that to Linux.
This isn't a new idea. Vertical wind turbines like that have been built before. They're not very good. A better vertical design is the Darrius parabolic vertical turbine. There used to be a few dozen of those at the Pacheco Pass wind farm, but they've been replaced with bladed units. Verticals have the advantage that all the equipment is at the bottom, but the side loads on the bearings are a big problem.
There are several hard problems in wind turbine design. One is that, for large wind machines, wind speed may vary considerably across different parts of the blade area. This produces huge stresses in the blade system. Aircraft propellers and hubs don't have that problem, so technology borrowed from aircraft props didn't quite work. That's been solved, but it took years to get past it.
A basic problem, one which this new design doesn't solve, is overspeed protection. Wind turbines above toy size must be able to deal with high wind conditions safely. Some turn sideways; some turn upwards; some feather the props. Brakes aren't enough. There's no way to feather or turn this new design. Even small turbines need, and have, overspeed protection.
There are lots of wind machine designs that more or less work in a small size, but don't scale up to the point where they're worth building. There's a square law; double the blade length and get four times the energy out. So big turbines beat out little ones, once ths scaling problems are solved. Wind turbine size has been creeping up since the 1970s, from about 50KW to a few megawatts.
A 1.5 MW unit was built in the 1940s, but it suffered a bearing failure within a year, then a loss of blade accident which threw a blade 700 feet. Only in the past decade have reliable wind machines in that size range been produced in quantity. With 2800 of their 1.5MW units installed, General Electric can be said to have solved that scaling problem.
The big machines aren't simple. They have active yaw control, active pitch control, hydraulic brakes, AC to DC to AC variable frequency conversion, and lightning protection. But, at last, they work.
So these guys are going to beat that with a little tin model that looks like something used to spin a sign in a used car lot. Right.
GeForce 6800 GS - $249, according to NVidia.
GeForce 6800 GT - $266, according to PriceGrabber.
The cheaper model has 12 instead of 16 pixel shaders, and 5 instead of 6 vertex shaders. They probably use the same chip. The benchmarks are close. $17 cheaper. Big deal.
In terms of price/performance, Via is probably the leader. They've just introduced some new S3 Chrome boards that are roughly comparable to the GEForce 6800 line, but are priced around $150. That technology will probably be in Via's motherboard chipsets soon, at an even lower price.
From the article:
Rick Maas, a chemist at the University of North Carolina at Asheville (UNC) who specialises in sustainable energy sources...
From the UNC catalog:
Richard Preston Maas (1987) Professor of Environmental Studies
B.A., Bucknell University; M.S., Western Carolina University; M.S.P.H., Ph.D., University of North Carolina at Chapel Hill
So far, the first court ruling indicates that National Security Letters are unenforceable and that the law authorizing them is unconstitutional. The Government is appealing, and the case was heard by the Second Circuit this fall. A decision is pending.
If you receive one, you need to get legal advice before complying.
The proposed legislation to criminalize NSL noncompliance,
S.1680, has no cosponsors and isn't going anywhere.
The FBI can still go before a judge and get a subpoena, but that requires judicial authorization, and you can fight a subpoena in court if it's overreaching.
Actual photo link, without blogs, etc.
on
Yahoo's Geek Statue
·
· Score: 2, Informative
The public Internet wasn't the first try. Look what came before:
Mead Data Central, which ran Lexis/Nexis. Good info at high prices.
Networked BBS systems, including Usenet over UUCP. Text info at low prices. (Anyone remember The Well?)
QuantumLink, a 2D virtual world with avatars. (With Commodore 64 clients at 300 baud! What a cram job.)
Minitel, the French system with good phone directories and expensive data services. (France Telecom fully deployed Minitel service in the United States, with dial-in ports all over the US. Few Americans used it, but the ability to send messages to France at no extra cost was great for anyone who spoke French. The literary standard expected in online chat was quite high.)
GEnie, Prodigy, MCImail, etc., the first big closed systems. Widely used, but not very good. No interoperablity, a big problem.
AOL, of course, which predates the Internet and didn't originally connect to it.
The big push to interconnect first came from E-mail. Business to business E-mail was a huge pain when GEnie didn't talk to MCImail. Businesses insisted that their vendors get interoperablity working.
That's what finally made the competing services interconnect.
That story reads like it was planted to hype XBox demand. The Xbox 360 is not selling for $1500 on eBay. It's selling for $380. With the hard drive and accessories.
There are some sellers asking higher prices, but none of them have any bids. Sellers can ask any "reserve price" they want, but it's not a real price unless someone will pay it.
I'm not a big fan of "vim", but it has its uses. One is looking at very large log files. Its large-file startup time is quite good, and it doesn't choke if you feed it a 50MB file.
Read "Compressing the Kill Chain", from Airforce Magazine. "All that administrative data that we can transmit from machine to machine leaves the human in the loop free to do much more important things that the machines can't do--like not get shot."
That's just the beginning.
Once the technology developed in the DARPA Grand Challenge really gets rolling, which, according to the director of DARPA, will be in about five years, America's armies of killer robots will go into the field.
The Army calls this the "Future Combat Systems" program.
180solutions is not a perpetrator and you can't implicate them in this scheme.
A good prosecutor might be able to bring it off. The legal definition of a punishable conspiracy is generally that at least two people knowingly cooperate to commit a crime, while in addition at least one of them does some illegal act to commit the crime.
180Solutions is already being sued in a class action. From the complaint: "180Solutions pays its distributors, who are its agents, money each time they infect a computer with 180Solutions spyware". Recruitment of agents and payment can be sufficient to establish a conspiracy, especially if there's a history of illegal acts by recruited agents. Read more of that filing to see how the plaintiff describes how 180Solutions not only tolerates, but pays, agents who use illegal means to force the install of 180Solutions software.
Since this lawsuit was filed, 180Solutions claims to be mending its ways. However, they're still allowing their existing affiliates to distribute the old, spyware-stuffed version of their application unti the end of 2005, so they're not too serious about it.
In any case, ceasing criminal activity is not a defense to previous crimes.
Actually, that's not a bad dance job. Pay, benefits, reasonable hours. Ask any working dancer. It's a tough life, and you burn out young.
At the higher levels, the injury rate is very high. New York City Ballet used to have the highest workmens's compensation premium in the state.
The "robot touch avoidance" demo has been done before, several times, both with mechanical switches and a short-range microwave system. The IR distance measurement system came from a Stanford project in the 1970s.
Re:What does this mean for San Fran and SBC Park?
on
Ma Bell is Back
·
· Score: 1
We should lobby Caltrans to just put up signs that say "SF Baseball Park" and "SF Football Stadium". No more of this nonsense.
There used to be a scientific time sharing industry, with mainframe computer time rented by the minute. It's dead. Most commercial jobs you can do on PCs. If you have an ongoing need for more crunch power than that, you can get your own computing power, and it will be cheaper than renting it. The market for huge numbers of intermittent cycles is weak to nonexistent. The basic problem is that there just aren't many companies with giant number-crunching jobs for which they are willing to pay. For the same reason, there are very few privately owned supercomputers. There was a "grid computing" utility about two years ago, before Sun tried it, and they didn't get customers either.
Sun's "grid computing" operation seems to be an attempt to find a use for unsold Sun servers, or at least to avoid writing their value down to scrap prices.
f you went to a big hosting company and said you wanted a thousand unlimited-CPU-at-low-priority shared hosting accounts, valid only from 2300 to 0700, you could probably get a really good price.
If "grid computing" were useful, somebody would be doing this. All those nearly idle CPUs could be doing something.
There's a successful grid computing company: Akamai. What they sell is distributed hosting and cacheing, which they call "Akamai On Demand Managed Services". When the web site for the World Cup or NASCAR or Britney is getting millions of hits per hour during some special event, thousands of Akamai servers switch to serving those pages to handle the transient load. That's a successful "grid" application, and it's been working for years.
Akamai does more than serve pages. You can run your business logic, in Java, on their servers. So they're already set up to run user code on their grid.
If anybody is going to sell grid computing profitably, it's Akamai. They're all set up to do it. Yet they don't.
There are ruggedized phones. The Motorola i325IS is qualified to MIL-STD 810F for ruggedness and waterproofing, and is approved as intrinsically safe for use in explosive atmospheres. Built-in off-network walkie talkie mode and GPS receiver. Usually used with Nextel, but available for most GSM-type services.
So if you actually need a ruggedized phone, they're available.
QNX definately rocks. Its what the Hurd wishes it could be.
Very true. The Hurd debacle is sad. QNX is good, but the company behind it has backed off from their "open QNX" effort, with an accompanying decline in market share.
Slashdot Mirror
The cell processors can do DMA to and from main memory while computing. As IBM puts it, "The most productive SPE memory-access model appears to be the one in which a list (such as a scatter-gather list) of DMA transfers is constructed in an SPE's local store so that the SPE's DMA controller can process the list asynchronously while the SPE operates on previously transferred data." So the cell processors basically have to be used as pipeline elements in a messaging system.
That's a tough design constraint. It's fine for low-interaction problems like cryptanalysis. It's OK for signal processing. It may or may not be good for rendering; the cell processors don't have enough memory to store a whole frame, or even a big chunk of one.
This is actually an old supercomputer design trick. In the supercomputer world, it was not too successful; look up the the nCube and the BBN Butterfly, all of which were a bunch of non-shared-memory machines tied to a control CPU. But the problem was that those machines were intended for heavy number-crunching on big problems, and those problems didn't break up well.
The closest machine architecturally to the "cell" processor is the Sony PS2. The PS2 is basically a rather slow general purpose CPU and two fast vector units. Initial programmer reaction to the PS2 was quite negative, and early games weren't very good. It took about two years before people figured out how to program the beast effectively. It was worth it because there were enough PS2s in the world to justify the programming headaches.
The small memory per cell processor is going to a big hassle for rendering. GPUs today let the pixel processors get at the frame buffer, dealing with the latency problem by having lots of pixel processors. The PS2 has a GS unit which owns the frame buffer and does the per-pixel updates. It looks like the cell architecture must do all frame buffer operations in the main CPU, which will bottleneck the graphics pipeline. For the "cell" scheme to succeed in graphics, there's going to have to be some kind of pixel-level GPU bolted on somewhere.
It's not really clear what the "cell" processors are for. They're fine for audio processing, but seem to be overkill for that alone. The memory limitations make them underpowered for rendering. And they're a pain to program for more general applications. Multicore shared-memory multiprocessors with good cacheing look like a better bet.
Read the cell architecture manual.
What they're actually talking about is the NetIntercept Appliance from Sandstorm Enterprises. This is also the FBI's replacement for Carnivore.
Microsoft may be able to succeed in this by rigging their OS so that it only works right with their ASP systems. But that didn't work for MSN. Nor was the Microsoft Connected Services Framework a success.
Yes, sometimes news breaks on a blog, but mostly, it's just drivel.
Now that Motorola has the hardware working, they can consider cutting Apple out of the loop. By, say, cutting a deal with WalMart.
Did the kernel buffer overflow in the .BMP/.RLE decoder ever get fixed? I was amazed to find that code in the NT/Win2K kernel.
The University of California at Berkeley won't accept for credit high school biology courses that teach intelligent design. If you want to get into the life sciences or medicine, get out of Kansas schools.
Actually, all drivers should be outside the kernel, as in QNX, and now Minix 3. But it's probably too late to do that to Linux.
There are several hard problems in wind turbine design. One is that, for large wind machines, wind speed may vary considerably across different parts of the blade area. This produces huge stresses in the blade system. Aircraft propellers and hubs don't have that problem, so technology borrowed from aircraft props didn't quite work. That's been solved, but it took years to get past it.
A basic problem, one which this new design doesn't solve, is overspeed protection. Wind turbines above toy size must be able to deal with high wind conditions safely. Some turn sideways; some turn upwards; some feather the props. Brakes aren't enough. There's no way to feather or turn this new design. Even small turbines need, and have, overspeed protection.
There are lots of wind machine designs that more or less work in a small size, but don't scale up to the point where they're worth building. There's a square law; double the blade length and get four times the energy out. So big turbines beat out little ones, once ths scaling problems are solved. Wind turbine size has been creeping up since the 1970s, from about 50KW to a few megawatts.
A 1.5 MW unit was built in the 1940s, but it suffered a bearing failure within a year, then a loss of blade accident which threw a blade 700 feet. Only in the past decade have reliable wind machines in that size range been produced in quantity. With 2800 of their 1.5MW units installed, General Electric can be said to have solved that scaling problem.
The big machines aren't simple. They have active yaw control, active pitch control, hydraulic brakes, AC to DC to AC variable frequency conversion, and lightning protection. But, at last, they work.
So these guys are going to beat that with a little tin model that looks like something used to spin a sign in a used car lot. Right.
GeForce 6800 GT - $266, according to PriceGrabber.
The cheaper model has 12 instead of 16 pixel shaders, and 5 instead of 6 vertex shaders. They probably use the same chip. The benchmarks are close. $17 cheaper. Big deal.
In terms of price/performance, Via is probably the leader. They've just introduced some new S3 Chrome boards that are roughly comparable to the GEForce 6800 line, but are priced around $150. That technology will probably be in Via's motherboard chipsets soon, at an even lower price.
Rick Maas, a chemist at the University of North Carolina at Asheville (UNC) who specialises in sustainable energy sources...
From the UNC catalog:
Richard Preston Maas (1987) Professor of Environmental Studies B.A., Bucknell University; M.S., Western Carolina University; M.S.P.H., Ph.D., University of North Carolina at Chapel Hill
"M.S.P.H." is "Master of Science in Public Health". His field is water quality. He's been an expert witness on lead leaching from bronze parts of water systems.
And where are the "65 peer reviewed papers"?
If you receive one, you need to get legal advice before complying.
The proposed legislation to criminalize NSL noncompliance, S.1680, has no cosponsors and isn't going anywhere.
The FBI can still go before a judge and get a subpoena, but that requires judicial authorization, and you can fight a subpoena in court if it's overreaching.
Direct link to actual photo.
The big push to interconnect first came from E-mail. Business to business E-mail was a huge pain when GEnie didn't talk to MCImail. Businesses insisted that their vendors get interoperablity working. That's what finally made the competing services interconnect.
There are some sellers asking higher prices, but none of them have any bids. Sellers can ask any "reserve price" they want, but it's not a real price unless someone will pay it.
I'm not a big fan of "vim", but it has its uses. One is looking at very large log files. Its large-file startup time is quite good, and it doesn't choke if you feed it a 50MB file.
The most famous robot kill was on November 3, 2002, when a Predator UAV equipped with a Hellfire missile blew up "six suspected Al-Queda terrorists" in Yemen.
That's just the beginning. Once the technology developed in the DARPA Grand Challenge really gets rolling, which, according to the director of DARPA, will be in about five years, America's armies of killer robots will go into the field. The Army calls this the "Future Combat Systems" program.
A good prosecutor might be able to bring it off. The legal definition of a punishable conspiracy is generally that at least two people knowingly cooperate to commit a crime, while in addition at least one of them does some illegal act to commit the crime.
180Solutions is already being sued in a class action. From the complaint: "180Solutions pays its distributors, who are its agents, money each time they infect a computer with 180Solutions spyware". Recruitment of agents and payment can be sufficient to establish a conspiracy, especially if there's a history of illegal acts by recruited agents. Read more of that filing to see how the plaintiff describes how 180Solutions not only tolerates, but pays, agents who use illegal means to force the install of 180Solutions software.
Since this lawsuit was filed, 180Solutions claims to be mending its ways. However, they're still allowing their existing affiliates to distribute the old, spyware-stuffed version of their application unti the end of 2005, so they're not too serious about it.
In any case, ceasing criminal activity is not a defense to previous crimes.
Here's a tour. It's a huge installation. Not in bad shape for a bunker, but will need considerable work to be usable.
OpenOffice Draw is about where Visio was when Microsoft bought it. It's good enough for drawings with boxes, circles, and arrows.
Actually, that's not a bad dance job. Pay, benefits, reasonable hours. Ask any working dancer. It's a tough life, and you burn out young. At the higher levels, the injury rate is very high. New York City Ballet used to have the highest workmens's compensation premium in the state.
The "robot touch avoidance" demo has been done before, several times, both with mechanical switches and a short-range microwave system. The IR distance measurement system came from a Stanford project in the 1970s.
We should lobby Caltrans to just put up signs that say "SF Baseball Park" and "SF Football Stadium". No more of this nonsense.
Sun's "grid computing" operation seems to be an attempt to find a use for unsold Sun servers, or at least to avoid writing their value down to scrap prices.
f you went to a big hosting company and said you wanted a thousand unlimited-CPU-at-low-priority shared hosting accounts, valid only from 2300 to 0700, you could probably get a really good price. If "grid computing" were useful, somebody would be doing this. All those nearly idle CPUs could be doing something.
There's a successful grid computing company: Akamai. What they sell is distributed hosting and cacheing, which they call "Akamai On Demand Managed Services". When the web site for the World Cup or NASCAR or Britney is getting millions of hits per hour during some special event, thousands of Akamai servers switch to serving those pages to handle the transient load. That's a successful "grid" application, and it's been working for years.
Akamai does more than serve pages. You can run your business logic, in Java, on their servers. So they're already set up to run user code on their grid. If anybody is going to sell grid computing profitably, it's Akamai. They're all set up to do it. Yet they don't.
So if you actually need a ruggedized phone, they're available.
Very true. The Hurd debacle is sad. QNX is good, but the company behind it has backed off from their "open QNX" effort, with an accompanying decline in market share.