Normally, I'm a big fan of very brilliant people...
but then I didn't think it was possible to jump the shark in a motorized wheelchair.
Seriously, what would any civilization want from a messed up, polluted tiny little planet like Earth? Any resource we have, except our own stupidity, is more abundant elsewhere.
Well, just to clarify, just because the ship is stealthy doens't mean that the reaction in the west should be to produce their own stealth ships. You don't need to react by mimic.
Instead, find out what the ship is designed for, its purpose. Stealth is not a purpose. In this case, the purpose is coastal defence. It is designed to be a defence against invasion. The west may not need a ship with the same purpose.
It helps to put it in perspective. My understanding of this subject is that after being brutally invaded in the last world war, the former USSR put a lot of money into defense against invasion, including patrol boats (such as this steath boat), coastal SSM and gun batteries, a huge anti-shipping marine air force (i.e. very long range bombers with ASM) and the military build up of a ground based defence in depth (i.e. the former Warsaw Pact). Perhaps they went over-board, but then i'm basised because my home country wasn't invaded in the last war.
As for what happens when all ships are stealthy? Well, i'm guessing that there is as much money going into detecting steathly vehicles (subs, ships, tanks, planes and missles) as there is going into building them. It's a battle between offense and defence that is as old as human civilization. As steath techniques make it difficult to detect using current radars, defence establishments will build new steath detecting radars (different frequencies, more power), or perhaps lasers. A good example that this stealth ship is not the beginning of the next dreadnaught-race, submarines have been near invisable for decades, but navies just learn to deal with the threat and continue operating.
In conclusion, i just don't see a need for an immediate reaction to a small (less than 2000 tons) patrol craft. Perhaps countries more fearing of sea-based invasion (ok, the UK would be on my list here, but also Germany and Japan come to mind). And that's not guessing at who the invader would be, but just a list of countries that would venerable.
BTW, it's not Big Blue's dominator. This machine comes from NEC and i believe is based on the SX-5 supercomputer.
The NEC SX-5 has the fastest CPU's in the supercomputing world. The main innovator for NEC Tadashi Watanabe is known as the Seymour Cray of Japan. Currently Cray has an agreement with NEC to resell the NEC supers, which are one of the only parallel vector machines still being produced. Cray stopped producing it's own version of these classic large PVP the Cray T90, and is now concentrating on the parallel, multithreaded and smaller PVP machines. Which left a gap at the high end, so Cray made the resellers agreement with NEC about 8 months ago
This machine should be able to acheive a much higher percentage of peak performance in production codes because of the huge memory to CPU bandwidth and because of using a smaller number of very very fast CPU's, there should be less parallel sync overhead. This should reduce the non-parallel times (see Amdahls Law).
I've used AIX a fair bit in the past. Some points of difference.
First AIX has a great file system. That is to say that i can use the admin tools to increase the size of a file system on a live system. No reboots. Perhaps some of the Linux JFS implementations have this as well.
Second, i quite like SMIT. I'm a programmer, not a paid sys admin, but no one else had UNIX experience, so i was stuck with support. SMIT, the admin tool, has everything all in one interface. No graphical shine, just a simple to use interface. Plus TTY support for when i don't have X11 access.
As was mentioned before, not all the config on an AIX box is through flat files. Some information is in the ODM object database. This is scary shit to me, and would perfer not to have to make changes in ODM. If you do it wrong, you can hose the machine. I think that Linux using only flat files is an advantage.
Porting to AIX is a hassle. At one point, it was said that AIX stood for 'AIX aIn't uniX'. Some of the porting is just dealing with file system differences (i.e,/usr vs/opt), others deal with usually subtle API differences. I've been responsible most of the porting here, and AIX was always the most difficult compared to Solaris and HP-UX. Linux has an advantage here, but i think AIX was getting better (certainly compared with AIX from the mid-90's.)
Lastly, the biggest difference for me was with accounting. While we don't use any (for our testing / lab machines), AIX has a lot of built in accounting that Linux seems to lack. If IBM wants Linux to replace AIX, this will most likely be where the most effort would be spent. AIX would logically be based on the security and accounting model used in the mainframe OS/390. I can't remember all the details, but you could set quotas on disk, cpu, memory, number of processes, log-in times, etc, etc. I don't think Linux has this kind of accounting quantum. Most Linux users are just not used to thinking about putting that much restrictions on user accesses. But if you put yourself in the mindset of people that would use that accounting (think banks, airline reservation systems, government, stock markets, medical facilities, etc).
It is my assumption that MS wants to kill Quicktime. One way of doing this is to put the MediaPlayer on Linux first. Remember, in the recent decade, it's not always the best product that wins, just the first.
Plus they can get extra media attention for their products, because all the news sites will run "MS ports products to Linux" stories for a few days (i.e. every business loves free good news stories about themselves".)
Finally, it can make it appear that they want to help Linux, when in fact, it is more likely they just want to be the only media platform on Linux, which then becomes a standard that they control.
I have a technical question about neutrinos. For many years, we have tried to find the mass of the neutrino (which to me, implies that they do not move at the speed of light, because that would require infinite energy). But the mass has been very elusive to measure. Do you suppose that the mass of the neutrino is related to the velocity? If, at the usual velocities we see (or rather, don't see) neutrinos (say, 99.9% C), their mass is undetectable. But as they approach the speed of light, their energy increases. Because energy and mass are related, perhaps as their energy (velocity) increases, their mass increases. The reason we have a hard time finding the mass, is that the neutrino mass is only detectable at velocities from which we can not currently obtain accurate measurements. For example, if we could create neutrinos (say in an accelerator) at 99.99% C, and the mass of the neutrino becomes large enough to measure only as it reaches 99.999999% C, perhaps it would appear that the neutrino does not have a mass. Assume that very high and very low velocity neutrinos are less abundant than medium velocity neutrinos. (This could have other implications, such as the omega value for the closure of the universe, or the lack of neutrinos measured from the Sun.)
1) I don't think that the G4 will come to the iMac for a very long time.
2) You are comparing 2 chips that aren't even for sale yet. I'm not sure that i'd feel comfortable comparing the 2002 ford taurus to the 2005 Toyota Camry.
3) The Merced is not designed in any way to be an upgrade to an existing chip. It is so totally different, and needs a different motherboard. A Mac example would be the 68K to PPC transistion. At that time, you couldn't just change the chip either (but you could get a card that replaced part of the motherboard ASICs).
4) Motorola is doing ok, just not in personal computers. One could ask why the PPC failed in the mainstream, but if there are only 2 options, MacOS and now BeOS, it just won't be a big seller. Ask instead where Pink, Taligent, Solaris PPC, AIX and OS/2 went. I think that the PPC couldn't take off like it should have without something to run on it.
5) You are comparing the G4 and Merced, but really they are for very different areas of the marketplace. I don't think anyone is suggesting that the Merced is for personal computer like the G4 is. Merced is for very high end workstations, and honking big servers.
6) i should tell you that I too am a Mac person. I'll never buy Intel or MS, ever. So i understand some of your feelings, but i just thought that your points were a little off base.
I was just reviewing some of the docs that Intel and HP released yesterday concerning Merced, and i have read between the lines. Let me know if you think i'm off-base with these assumtions:
1) Merced will max out at 750Mhz in it's first implementation. In the 8 page Application Developer's Arch Gde, they state that Merced will acheive 6Gflops in it's first implementation, and that it is acheieved with 4 Multiply-Accumlate units (FMAC). Each unit can do 2 flops, so the chip can do 8 flops per clock. 6Gflops, divided by 8 flops per clock gives a clock cycle of 750Mhz. And IF the chip could go faster, i would think Intel would use those higher numbers. So i think it's safe to think that 750Mhz is all that Intel can produce at first.
2) Since the maximum Gflops rating would include all FMAC units (that's the assumtion i'm using), then that would indicate that the Mecrd chip will execute 2 LIW instructions per clock. Each LIW word has 3 instructions, which is not enough to send 4 FMAC instructions, so Merced must be able to send 2 of the 128-bit instruction words to the execution units.
3) Assuming #2 is right (it might not be), then Merced can perform 6 instructions (3 in each LIW word) per clock for a maximum MIPS of 4500 instructions per second (yes, yes, i know how useless MIPS is for benchmarking). Or, because of the FMAC units doing 2 ops, 10,000 Mops, or 10 billion operations per second.
I'll be reading the 400+ page Intel doc today, and i hope to find more information about memory access. The above rates assume that the chip can get enough bandwidth to use all the units above. Just the instructions alone mean that Merced needs 24Gbytes/second (256-bits per clock, 750Mhz) of cache/CPU bandwidth.
The ASCI RED machine is not at LLNL, but at Sandia. It does use PPros. But a parallel machine with P-II's is also possible. It just takes an investment in designing your own chip-set. The PPro chipset supports low order SMP natively in most chipsets. But for the RED machine, Intel used custom networking chipsets to get the large machine. This can be done with Pentium-II's if you wanted to do that. And a 149 processor Pentium-II would not kill a Cray. Well, not a large one at least. I would guess 149 processor P-II would come in around the same as a 100 processor Cray T3E or Origin2000. But since they are selling T3E's as large as 1380 processors, and Origins in the thousands of processors, a small 149 processor would not come close.
I've reviewed the information about HAL, and am very disappointed. There are misleading or just plain wrong information at the web site. I'm not saying that it's not a good machine,just that i would be careful drawing any real comparisons from the HAL documents.
1) If you read the "About Company" link, it states that in fact they have not actually built the machine that they claim will reach 12.8 Tops. It states that it will be built by Feb 99. I believe that they scaled the performance of a machine much much smaller than the one they claim betters the IBM machine. Note that scaling performance of a smaller machine is considered a big faux pas, as it totally neglects parallel overheads.
2) They quote 4-bit operations as though they are equivalent to the 32-bit Floating point ops that were tested on BlueMountain. They are not similar. The performance drops off to 3.8Tops when they use 16-bit operations, but the IBM is still doing twice as much real work because they are using 32-bit math.
3) They have not actually stated which benchmarks they used (if they actually had a machine to test). The IBM used (i think) the LinPeak, which is a matrix multiply operation. That benchmark is very good at showing off parallel architectures. But the size of the matrix must be quoted in the results. I don't see that information.
4) The HAL computer is very under memoried. This might be easy to fix. The rule of thumb is 1byte RAM per Flop. That's why the BLUE machine has 2.5Tbytes of RAM for 3.8TFlops. The max listed for HAL is 100GB, or about 1-2 orders of magnitude too small.
5) On one page at StarBridge, it lists the I/O performance in comparison to a Cray T3E-1024. But the numbers are differnt on the press release. In the press release is sez 50GB/s, but the company link indicates 50MB/s. 3 orders of magnitude different. I wonder if one of thoses is actual I/O data, and one is (50GB) is supposed to be memory bandwidth, which is a very very different concept from I/O. If the machine only has 50MB/s I/O, but 100GB of memory, then it will take about 2000 seconds to load memory from disk (or, say, to do a check point of the current data set).
6) The marketing data for the Cray T3E-1024 is wrong, which in my mind negates most of the comparisons. The Cray T3E-1024 does not cost $76m, but the Cray/SGI Blue Origin2000-3000 does cost about that much. It states the Cray does not have fault tolerence, but it most certainly does. It states that the maximum I/O of the T3E is less than 2GB/s, which i know is wrong. (Did someone in marketting write this without double checking the numbers??)
7) The company link states that the HAL 4rW1 has a minimum sustained rate of 3.8 Tera-ops. Would any company really claim that there is no program that they could run that would not perform worse. If that is a challenge, i'd be willing to put a huge bet down that i could write a program that will give LESS than 3.8 Tops!!!
8) programming. On the company page, it does not state that you can program this in C or Fortran or any other common language. Instead it only talks about the GUI that you can use to describe your problem, and the software will automagically start running (at 3.8Tops minimum!!). This is a HUGE drawback if this is true (no C or Fortran). Also how can they possibly claim that their machine does not have the same parallel exectution overheads that are common on other parallel machines?? Just because you can change your network on the fly (and just how long does that change take??) does not mean you are immune to Amdahl's law!!
So to sum up: 1) 4-bit math is NOT equivalent to 32-bit math. 2) What are the benchmarks and data sets?? What were they programmed with? Was the test actually run with a full scale machine, or just scaled results?? 3) How to deal with Amdahls law? 4) Hardware is under memoried, and very likely less than perfect I/O performance.
Slashdot Mirror
Normally, I'm a big fan of very brilliant people... but then I didn't think it was possible to jump the shark in a motorized wheelchair. Seriously, what would any civilization want from a messed up, polluted tiny little planet like Earth? Any resource we have, except our own stupidity, is more abundant elsewhere.
Nortel Networks already announced this type of program back in December 2002:
w sr eleases/2002d/12_03_02_wireless_lan.html
http://www.nortelnetworks.com/corporate/news/ne
(Safe Harbor: I work for Nortel; this is a public news release at the corporate web site).
Well, just to clarify, just because the ship is stealthy doens't mean that the reaction in the west should be to produce their own stealth ships. You don't need to react by mimic.
Instead, find out what the ship is designed for, its purpose. Stealth is not a purpose. In this case, the purpose is coastal defence. It is designed to be a defence against invasion. The west may not need a ship with the same purpose.
It helps to put it in perspective. My understanding of this subject is that after being brutally invaded in the last world war, the former USSR put a lot of money into defense against invasion, including patrol boats (such as this steath boat), coastal SSM and gun batteries, a huge anti-shipping marine air force (i.e. very long range bombers with ASM) and the military build up of a ground based defence in depth (i.e. the former Warsaw Pact). Perhaps they went over-board, but then i'm basised because my home country wasn't invaded in the last war.
As for what happens when all ships are stealthy? Well, i'm guessing that there is as much money going into detecting steathly vehicles (subs, ships, tanks, planes and missles) as there is going into building them. It's a battle between offense and defence that is as old as human civilization. As steath techniques make it difficult to detect using current radars, defence establishments will build new steath detecting radars (different frequencies, more power), or perhaps lasers. A good example that this stealth ship is not the beginning of the next dreadnaught-race, submarines have been near invisable for decades, but navies just learn to deal with the threat and continue operating.
In conclusion, i just don't see a need for an immediate reaction to a small (less than 2000 tons) patrol craft. Perhaps countries more fearing of sea-based invasion (ok, the UK would be on my list here, but also Germany and Japan come to mind). And that's not guessing at who the invader would be, but just a list of countries that would venerable.
BTW, it's not Big Blue's dominator. This machine comes from NEC and i believe is based on the SX-5 supercomputer.
The NEC SX-5 has the fastest CPU's in the supercomputing world. The main innovator for NEC Tadashi Watanabe is known as the Seymour Cray of Japan. Currently Cray has an agreement with NEC to resell the NEC supers, which are one of the only parallel vector machines still being produced. Cray stopped producing it's own version of these classic large PVP the Cray T90, and is now concentrating on the parallel, multithreaded and smaller PVP machines. Which left a gap at the high end, so Cray made the resellers agreement with NEC about 8 months ago
This machine should be able to acheive a much higher percentage of peak performance in production codes because of the huge memory to CPU bandwidth and because of using a smaller number of very very fast CPU's, there should be less parallel sync overhead. This should reduce the non-parallel times (see Amdahls Law).
I've used AIX a fair bit in the past. Some points of difference.
/usr vs /opt), others deal with usually subtle API differences. I've been responsible most of the porting here, and AIX was always the most difficult compared to Solaris and HP-UX. Linux has an advantage here, but i think AIX was getting better (certainly compared with AIX from the mid-90's.)
First AIX has a great file system. That is to say that i can use the admin tools to increase the size of a file system on a live system. No reboots. Perhaps some of the Linux JFS implementations have this as well.
Second, i quite like SMIT. I'm a programmer, not a paid sys admin, but no one else had UNIX experience, so i was stuck with support. SMIT, the admin tool, has everything all in one interface. No graphical shine, just a simple to use interface. Plus TTY support for when i don't have X11 access.
As was mentioned before, not all the config on an AIX box is through flat files. Some information is in the ODM object database. This is scary shit to me, and would perfer not to have to make changes in ODM. If you do it wrong, you can hose the machine. I think that Linux using only flat files is an advantage.
Porting to AIX is a hassle. At one point, it was said that AIX stood for 'AIX aIn't uniX'. Some of the porting is just dealing with file system differences (i.e,
Lastly, the biggest difference for me was with accounting. While we don't use any (for our testing / lab machines), AIX has a lot of built in accounting that Linux seems to lack. If IBM wants Linux to replace AIX, this will most likely be where the most effort would be spent. AIX would logically be based on the security and accounting model used in the mainframe OS/390. I can't remember all the details, but you could set quotas on disk, cpu, memory, number of processes, log-in times, etc, etc. I don't think Linux has this kind of accounting quantum. Most Linux users are just not used to thinking about putting that much restrictions on user accesses. But if you put yourself in the mindset of people that would use that accounting (think banks, airline reservation systems, government, stock markets, medical facilities, etc).
Plus they can get extra media attention for their products, because all the news sites will run "MS ports products to Linux" stories for a few days (i.e. every business loves free good news stories about themselves".)
Finally, it can make it appear that they want to help Linux, when in fact, it is more likely they just want to be the only media platform on Linux, which then becomes a standard that they control.
I have a technical question about neutrinos. For many years, we have tried to find the mass of the neutrino (which to me, implies that they do not move at the speed of light, because that would require infinite energy). But the mass has been very elusive to measure. Do you suppose that the mass of the neutrino is related to the velocity? If, at the usual velocities we see (or rather, don't see) neutrinos (say, 99.9% C), their mass is undetectable. But as they approach the speed of light, their energy increases. Because energy and mass are related, perhaps as their energy (velocity) increases, their mass increases. The reason we have a hard time finding the mass, is that the neutrino mass is only detectable at velocities from which we can not currently obtain accurate measurements. For example, if we could create neutrinos (say in an accelerator) at 99.99% C, and the mass of the neutrino becomes large enough to measure only as it reaches 99.999999% C, perhaps it would appear that the neutrino does not have a mass. Assume that very high and very low velocity neutrinos are less abundant than medium velocity neutrinos. (This could have other implications, such as the omega value for the closure of the universe, or the lack of neutrinos measured from the Sun.)
1) I don't think that the G4 will come to the iMac for a very long time.
2) You are comparing 2 chips that aren't even for sale yet. I'm not sure that i'd feel comfortable comparing the 2002 ford taurus to the 2005 Toyota Camry.
3) The Merced is not designed in any way to be an upgrade to an existing chip. It is so totally different, and needs a different motherboard. A Mac example would be the 68K to PPC transistion. At that time, you couldn't just change the chip either (but you could get a card that replaced part of the motherboard ASICs).
4) Motorola is doing ok, just not in personal computers. One could ask why the PPC failed in the mainstream, but if there are only 2 options, MacOS and now BeOS, it just won't be a big seller. Ask instead where Pink, Taligent, Solaris PPC, AIX and OS/2 went. I think that the PPC couldn't take off like it should have without something to run on it.
5) You are comparing the G4 and Merced, but really they are for very different areas of the marketplace. I don't think anyone is suggesting that the Merced is for personal computer like the G4 is. Merced is for very high end workstations, and honking big servers.
6) i should tell you that I too am a Mac person. I'll never buy Intel or MS, ever. So i understand some of your feelings, but i just thought that your points were a little off base.
-r.
I was just reviewing some of the docs that Intel and HP released
yesterday concerning Merced, and i have read between the lines. Let me
know if you think i'm off-base with these assumtions:
1) Merced will max out at 750Mhz in it's first implementation. In the 8
page Application Developer's Arch Gde, they state that Merced will
acheive 6Gflops in it's first implementation, and that it is acheieved
with 4 Multiply-Accumlate units (FMAC). Each unit can do 2 flops, so the
chip can do 8 flops per clock. 6Gflops, divided by 8 flops per clock
gives a clock cycle of 750Mhz. And IF the chip could go faster, i would
think Intel would use those higher numbers. So i think it's safe to
think that 750Mhz is all that Intel can produce at first.
2) Since the maximum Gflops rating would include all FMAC units (that's
the assumtion i'm using), then that would indicate that the Mecrd chip
will execute 2 LIW instructions per clock. Each LIW word has 3
instructions, which is not enough to send 4 FMAC instructions, so Merced
must be able to send 2 of the 128-bit instruction words to the execution
units.
3) Assuming #2 is right (it might not be), then Merced can perform 6
instructions (3 in each LIW word) per clock for a maximum MIPS of 4500
instructions per second (yes, yes, i know how useless MIPS is for
benchmarking). Or, because of the FMAC units doing 2 ops, 10,000 Mops,
or 10 billion operations per second.
I'll be reading the 400+ page Intel doc today, and i hope to find more
information about memory access. The above rates assume that the chip
can get enough bandwidth to use all the units above. Just the
instructions alone mean that Merced needs 24Gbytes/second (256-bits per
clock, 750Mhz) of cache/CPU bandwidth.
The ASCI RED machine is not at LLNL, but at Sandia. It does use PPros. But a parallel machine with P-II's is also possible. It just takes an investment in designing your own chip-set. The PPro chipset supports low order SMP natively in most chipsets. But for the RED machine, Intel used custom networking chipsets to get the large machine. This can be done with Pentium-II's if you wanted to do that.
And a 149 processor Pentium-II would not kill a Cray. Well, not a large one at least. I would guess 149 processor P-II would come in around the same as a 100 processor Cray T3E or Origin2000. But since they are selling T3E's as large as 1380 processors, and Origins in the thousands of processors, a small 149 processor would not come close.
-r.
I've reviewed the information about HAL, and am very disappointed. There are misleading or just plain wrong information at the web site. I'm not saying that it's not a good machine,just that i would be careful drawing any real comparisons from the HAL documents.
1) If you read the "About Company" link, it states that in fact they have not actually built the machine that they claim will reach 12.8 Tops. It states that it will be built by Feb 99. I believe that they scaled the performance of a machine much much smaller than the one they claim betters the IBM machine. Note that scaling performance of a smaller machine is considered a big faux pas, as it totally neglects parallel overheads.
2) They quote 4-bit operations as though they are equivalent to the 32-bit Floating point ops that were tested on BlueMountain. They are not similar. The performance drops off to 3.8Tops when they use 16-bit operations, but the IBM is still doing twice as much real work because they are using 32-bit math.
3) They have not actually stated which benchmarks they used (if they actually had a machine to test). The IBM used (i think) the LinPeak, which is a matrix multiply operation. That benchmark is very good at showing off parallel architectures. But the size of the matrix must be quoted in the results. I don't see that information.
4) The HAL computer is very under memoried. This might be easy to fix. The rule of thumb is 1byte RAM per Flop. That's why the BLUE machine has 2.5Tbytes of RAM for 3.8TFlops. The max listed for HAL is 100GB, or about 1-2 orders of magnitude too small.
5) On one page at StarBridge, it lists the I/O performance in comparison to a Cray T3E-1024. But the numbers are differnt on the press release. In the press release is sez 50GB/s, but the company link indicates 50MB/s. 3 orders of magnitude different. I wonder if one of thoses is actual I/O data, and one is (50GB) is supposed to be memory bandwidth, which is a very very different concept from I/O. If the machine only has 50MB/s I/O, but 100GB of memory, then it will take about 2000 seconds to load memory from disk (or, say, to do a check point of the current data set).
6) The marketing data for the Cray T3E-1024 is wrong, which in my mind negates most of the comparisons. The Cray T3E-1024 does not cost $76m, but the Cray/SGI Blue Origin2000-3000 does cost about that much. It states the Cray does not have fault tolerence, but it most certainly does. It states that the maximum I/O of the T3E is less than 2GB/s, which i know is wrong. (Did someone in marketting write this without double checking the numbers??)
7) The company link states that the HAL 4rW1 has a minimum sustained rate of 3.8 Tera-ops. Would any company really claim that there is no program that they could run that would not perform worse. If that is a challenge, i'd be willing to put a huge bet down that i could write a program that will give LESS than 3.8 Tops!!!
8) programming. On the company page, it does not state that you can program this in C or Fortran or any other common language. Instead it only talks about the GUI that you can use to describe your problem, and the software will automagically start running (at 3.8Tops minimum!!). This is a HUGE drawback if this is true (no C or Fortran). Also how can they possibly claim that their machine does not have the same parallel exectution overheads that are common on other parallel machines?? Just because you can change your network on the fly (and just how long does that change take??) does not mean you are immune to Amdahl's law!!
So to sum up:
1) 4-bit math is NOT equivalent to 32-bit math.
2) What are the benchmarks and data sets?? What were they programmed with? Was the test actually run with a full scale machine, or just scaled results??
3) How to deal with Amdahls law?
4) Hardware is under memoried, and very likely less than perfect I/O performance.
-r.