This is the precise difference between the commercial business model and the free software model. IMNSHO, any business that can only thrive when their customers need them deserves extinction.
??? My head is hurting. Businesses that thrive when their customers need them deserve extinction? If customers _don't_ need a business then that business won't exist anyone. Are you saying business don't deserve to exist at all.
The way things work is that businesses exist soley because they're providing things customers need.
Yes, some people do budget that way. If you're like me, and running large cluster computers, then higher electricity consumption means (tens of) thousands of dollars in electricity charges, and a requirement for bigger, more expensive airconditioning. Those expenses aren't trivial.
Similarily for large corporation deploying 30,000 new workstations, the extra electrical charges add up to a big number.
1. US tax dollars aren't doing much to keep the India/Pakistan. There is a UN observer force in Kashmir observing the adherence of the two countries to their ceasefire, but no peace-keeping force preventing another war over the zone. And since the US doesn't pay it's dues, it's not really even funding that small group
3. The US dollar is not artificially high. By and large currencies float on free market exchange. Cental banks typically have little control over this.
Take the Canadian dollar for instance. Each day, the amount of trading in the Canadian dollar is over a trillion dollars. The Bank of Canada has in the order of $50B in reserves. If the currency markets lose faith in the dollar, there is nothing the bank can do to prop it up.
The situation is the same for every country who floats their currency (including the US). As a matter of fact, the US dollar has dipped in the past year or two due to its faltering economy.
The rest of points need arguing with too, leave your email address and I'll be happy to continue
SPARC might be OK at high-throughput jobs, but IA32 and PowerPC just smash it to little bits for things that are less sequential.
Huh? Sparc (and Solaris) are specifically designed to run "less sequential" (ie, easily parallelizable) code. SPARC clock speeds are poor individually, but put 72 of them in a box and you can do things no IA32 machine can.
What single processor would be equivalent to a cluster of 3 P2 300MHz?
May a 500-600 MHz CPU. Tops. Seriously. Mainly because of the following - that old computer probably has a 10Mbit NIC, maybe 100Mbit. Can you say latency?
If the game server you want to run is multi-threaded then you _might_ be able to run different threads on different nodes (using say OpenMosix). It'll probably be slow as hell because of the latency. Probably slower than running it on machine.
Look, clusters are good for running parallel applications. Nothing else. If you don't have any parallel applications don't bother. It's not going to speed up your game server, or your game client, or your OS.
And look at it this way too, it costs around $150-200 a year in electricity to run a computer. You could buy a single CPU machine that's faster than the combined speed of your three computers and still save money because of electricity.
Save something for the next fifty or sixty years of your life.
Don't _EVER_ save anything for later that you can today. Because you may never get another chance. Don't rush out blindly and be reckless about things, but don't put them off purposefully.
Electricity costs money, often lot's of it, therefore flops/watt is a very important measure.
On the other hand, the article does make an error, it shoud be "one billon floating point operations per watt" or 1GFlop/watt. As opposed to an Itanium which gets about 0.05GFlop/watt.
Ummm...no, not very wrong. One dual-xeon server is ~$2.5K, four dual-xeon servers is ~$10K.
Bigger SMP machines, ie 4-ways, 8-ways, 20-ways, 106-ways cost considerably more. I know that, I have a 20-CPUs Sunfire 10feet behind me. But those aren't the computers I was talking about.
The Orion isn't a 12-way machine. It's 12 different machines in one chassis. I was comparing it to the normal equivalent in the clustering world. Which these days are typically dual-cpu compute nodes.
12-way CPU I could be wrong, but in computing "12-way CPU" implies a single system image, shared memory computer running across 12 CPUS. The Orion system is running a separate OS instance on each CPU, each with it's own local memory and communicating via ethernet.
The Orion in 12 1-way systems in a single chassis.
The 12-CPU system here scored 18GFlops sustained (using Linpack), 36GFlops peak. That's 1.5/3 GFlops. The CPUs are 1.4GHz, so that's 2.15flops/cycle.
By contrast, Xeons are 2flops/cycles. So a 3GHz Xeon can achieve 6GHz peak. Given the efficiency of the processor, and inefficiences from parallelizing,expected performance is probably around 3-4 GFlops. A dual 3GHz server runs about $2500, so that means 8 Xeons for about the $10K price of the Orion. So that's 24-32Gflops expected, 48Gflops peak.
Based strictly on performance, the Xeons have the edge, however
1.Space - A a company doing rendering could stick a couple of these under desks instead of allocating an entire room for computing (space costs money).
2.Electrical - The Orion uses the same amount of power as just one dual-Xeon node. The extra nodes will cost about $400-500 in electricity each year. Ammortized across a 3-year system lifespan that's $1500.
3.A/C - The Orion puts off as a much heat as a desktop. No need to get massive (expensive) A/C units to pump tons of cold air into a server room.
Prices aren't going to sink to the lowest. Nobody with the slightest comprehension of economics would suggest that. Prices will find some middle ground, rules of supply and demand.
Take the US and India for example. A US worker makes say $120K, the same worker in India might make $5K. Demand for the Indian labour will increase, therefore price will increase, vice versa in the US.
And it's true. Wages in the US/Canada tech sector aren't what they were 5 years. New grads aren't making $80K out school anymore. At the same time salaries in India have gone up.
Globalism evens out the spread of wealth, and theoretically the emergance of middle-classes in developing countries will create enough new wealth to offset the losses initially felt in wealthier countries.
If the scientists are influential, then they're important. The fact that their insights are being followed and expanded upon by others makes them important.
And it's a pretty good guess that they're competent too. If they weren't good scientists then why would their colleages be influenced by them?
It's a perfectly valid quote. How is it "pretty odd" or "unfortunate"?. How does it imply that Redhat doesn't believe in its business model?
Redhat's business model is:
Take an available open-source OS
Use it's own engineers to build upon the OS
Sell the OS into the market
That's they're business model and they aren't changing it. They're just focusing their market. In the desktop market, parts 1&2 don't allow them to compete with MS. The features needed to compete don't exist in the OS, and RH doesn't have the man-power to add them in itself.
In the server market, which RH is currently targetting, they can very effectively compete. So RH is expending it's limited resources to help in compete in this market instead.
there's this stereotype that all asian men are desperate people. The entire culture is surrounded about finding/having boyfriends/girlfriends. Every track on every asian CD, especially the ones you find in hong kong have this same theme, and every TV "drama" has the same themes. In both highschool and college, it seems that meeting someone is their ONLY purpose of existence. When you meet up with friends or family relative members that you haven't seen in a long while, 3 sentences within the conversation is ALWAYS whether you have a boyfriend/girlfriend yet.
Because the customer reps lie, lie, lie. EVERY time I've had a rep try to sell me on the service plan I'm told it covers accidental damage. Contrary to the fine print two pages into the product service plan details.
And since the product service plan _itself_ doesn't need to be signed, just the receipt, the employee often keeps the service plan in their hands while completing the sale. The employee thus shows only the side with the big point form friendly details not the fine print.
It isn't unreasonable that people are deceived into buying the PSP, and it's rather bigoted to claim they're morons.
1. Instead of offering a $20 rebate and redeeming 50% of those, why not just offere a straight at the counter $10 discount. Pisses off less people
2. I bought a TV-tuner card a couple years ago because it advertised a $50 rebate. Buy it, open it up to read to rebate information inside the box only to find out the rebate expired months before I purchased it. Was I pissed off? Damn straight.
But if the supercomputer is more efficient per raw unit of power, then the price per unit doesn't matter.
I work for living with HPC, buth with clusters and with large SMP machines. The cluster is nice, but there are some things than can _only_ be run a large SMP machine or are much, much faster on a SMP.
Right and no. If engineers/scientists need to wait for results, then two things happen:
1. Run a cluster and wait two weeks. Twiddle thumbs in mean time.
2. Run on supercomputer. Start processing next day.
For a lot of companies, it is justifiable to spend more money to acheive #2.
Not too mention that a lot of large programs that can't be parallelized have high memory requirements. If your program needs 50GB of memory and can't be parallelized to run across a cluster then you have no choice but run on a supercomputer, since no cheap system can handle 50GB.
No, they don't (not really anyway). Google isn't a HPC cluster by the way.
Take the big 1000+ CPU clusters build by labs and universities. They aren't running the same application parallelized across 1000 nodes. What happens is UserA gets 16 CPUs, UserB gets 32, maybe 10 for UserC.
Most calculations don't scale well beyond a certain number of processors, and often adding more CPUs will _slow_ down performance. Depending on the problem, a large supercomputer can scale better than a cluster.
Some applications are parallelized to run on 500, 1000, more nodes but those are very rare and very hard to write.
1. Cray is definitely pro-linux. It's what their XD1 runs. Though not their bigger computers.
2. There are some problems for which that a cluster can not even come close to achieving the performance of a supercomputer. For a lot of problems yes, for some maybe if you spend a fortune on fancy interconnects, and for some no.
3. If you're commercially building clusters let me know company it is. I'm in the market for a 128CPU cluster and I want to know who not to buy from.
There's an entire branch of parallel application which are labeled "embarrassingly parallel". This description simply means that such programs are trivially parallelized and achieve as close to linear as possible when scaled across many nodes. This is because of the low inter-node communications.
For "embarrassingly parallel" applications, a cluster is a really good tool. For programs that parallelize as nicely a nice big vector or smp will do nicely. Some code will run better on small 20CPU SMP machine than on a 1000 node cluster.
1) Company A raises $100M through debt-financing, talking a 20year loan at a 5% interest compounded annually. Over the 20 years has an average pre-debt-payment profit of $8M. Given yearly debt-payments of $6M, this leaves $2M/year for the company founder, who then has a net worth at the end of 20 years ~$40M.
2) Company B raises $100M through an IPO, with the founder retaining 20% of shares. Given the same profit, the company is free to give out say $6M in dividends (take off some from taxes). The founder then receives $1.2M for a total of $24M at the end of 20 years.
So it appears you're right then and the IPO way isn't as good....right? But wait, the founder still has stocks initially worth $20M which now puts him out ahead. But, then consider that given a standard P/E ratio of 30 the market capitalization for the company is ~$240M, making the founders stake worth $48M, giving him a total worth of $72M.
Of course there's a whole host of other things that affect things one way or the other. Like personal income taxes. Founder A is paying tax on $40M, while Founder B pays on only $24M, since stocks aren't taxed until you sell of them.
Before there was Intel x86 (comptabile) and a number of niche processors, and now there's still Intel and a number of niche processors. The submitter's closing statement seems a tad alarmist.
We still have Itanium, two Sparc variants, a number of Power variants, Transmeta, Opteron, and whole bunch of other niche processors, most of which probably have more market share than alpha.
1. Yes, in fact you are debt free. No HAH's about it.
2. (almost) No stockholder is going to purposely devalue stock they own just because you don't obey their whims. They are in this to make money afterall, not to megalomaniacally micromanage comparies.
3. Once that $50M is raised initially the stock could go down to $0, and it wouldn't change a thing.
Slashdot Mirror
This is the precise difference between the commercial business model and the free software model. IMNSHO, any business that can only thrive when their customers need them deserves extinction.
??? My head is hurting. Businesses that thrive when their customers need them deserve extinction? If customers _don't_ need a business then that business won't exist anyone. Are you saying business don't deserve to exist at all.
The way things work is that businesses exist soley because they're providing things customers need.
Yes, some people do budget that way. If you're like me, and running large cluster computers, then higher electricity consumption means (tens of) thousands of dollars in electricity charges, and a requirement for bigger, more expensive airconditioning. Those expenses aren't trivial.
Similarily for large corporation deploying 30,000 new workstations, the extra electrical charges add up to a big number.
Regarding your points
1. US tax dollars aren't doing much to keep the India/Pakistan. There is a UN observer force in Kashmir observing the adherence of the two countries to their ceasefire, but no peace-keeping force preventing another war over the zone. And since the US doesn't pay it's dues, it's not really even funding that small group
3. The US dollar is not artificially high. By and large currencies float on free market exchange. Cental banks typically have little control over this.
Take the Canadian dollar for instance. Each day, the amount of trading in the Canadian dollar is over a trillion dollars. The Bank of Canada has in the order of $50B in reserves. If the currency markets lose faith in the dollar, there is nothing the bank can do to prop it up.
The situation is the same for every country who floats their currency (including the US). As a matter of fact, the US dollar has dipped in the past year or two due to its faltering economy.
The rest of points need arguing with too, leave your email address and I'll be happy to continue
SPARC might be OK at high-throughput jobs, but IA32 and PowerPC just smash it to little bits for things that are less sequential.
Huh? Sparc (and Solaris) are specifically designed to run "less sequential" (ie, easily parallelizable) code. SPARC clock speeds are poor individually, but put 72 of them in a box and you can do things no IA32 machine can.
What single processor would be equivalent to a cluster of 3 P2 300MHz?
May a 500-600 MHz CPU. Tops. Seriously. Mainly because of the following - that old computer probably has a 10Mbit NIC, maybe 100Mbit. Can you say latency?
If the game server you want to run is multi-threaded then you _might_ be able to run different threads on different nodes (using say OpenMosix). It'll probably be slow as hell because of the latency. Probably slower than running it on machine.
Look, clusters are good for running parallel applications. Nothing else. If you don't have any parallel applications don't bother. It's not going to speed up your game server, or your game client, or your OS.
And look at it this way too, it costs around $150-200 a year in electricity to run a computer. You could buy a single CPU machine that's faster than the combined speed of your three computers and still save money because of electricity.
Save something for the next fifty or sixty years of your life.
Don't _EVER_ save anything for later that you can today. Because you may never get another chance. Don't rush out blindly and be reckless about things, but don't put them off purposefully.
"one floating point operation per watt"
Electricity costs money, often lot's of it, therefore flops/watt is a very important measure.
On the other hand, the article does make an error, it shoud be "one billon floating point operations per watt" or 1GFlop/watt. As opposed to an Itanium which gets about 0.05GFlop/watt.
Ummm...no, not very wrong. One dual-xeon server is ~$2.5K, four dual-xeon servers is ~$10K.
Bigger SMP machines, ie 4-ways, 8-ways, 20-ways, 106-ways cost considerably more. I know that, I have a 20-CPUs Sunfire 10feet behind me. But those aren't the computers I was talking about.
The Orion isn't a 12-way machine. It's 12 different machines in one chassis. I was comparing it to the normal equivalent in the clustering world. Which these days are typically dual-cpu compute nodes.
12-way CPU
I could be wrong, but in computing "12-way CPU" implies a single system image, shared memory computer running across 12 CPUS. The Orion system is running a separate OS instance on each CPU, each with it's own local memory and communicating via ethernet.
The Orion in 12 1-way systems in a single chassis.
The 12-CPU system here scored 18GFlops sustained (using Linpack), 36GFlops peak. That's 1.5/3 GFlops. The CPUs are 1.4GHz, so that's 2.15flops/cycle.
By contrast, Xeons are 2flops/cycles. So a 3GHz Xeon can achieve 6GHz peak. Given the efficiency of the processor, and inefficiences from parallelizing,expected performance is probably around 3-4 GFlops.
A dual 3GHz server runs about $2500, so that means 8 Xeons for about the $10K price of the Orion. So that's 24-32Gflops expected, 48Gflops peak.
Based strictly on performance, the Xeons have the edge, however
1.Space - A a company doing rendering could stick a couple of these under desks instead of allocating an entire room for computing (space costs money).
2.Electrical - The Orion uses the same amount of power as just one dual-Xeon node. The extra nodes will cost about $400-500 in electricity each year. Ammortized across a 3-year system lifespan that's $1500.
3.A/C - The Orion puts off as a much heat as a desktop. No need to get massive (expensive) A/C units to pump tons of cold air into a server room.
Prices aren't going to sink to the lowest. Nobody with the slightest comprehension of economics would suggest that. Prices will find some middle ground, rules of supply and demand.
Take the US and India for example. A US worker makes say $120K, the same worker in India might make $5K. Demand for the Indian labour will increase, therefore price will increase, vice versa in the US.
And it's true. Wages in the US/Canada tech sector aren't what they were 5 years. New grads aren't making $80K out school anymore. At the same time salaries in India have gone up.
Globalism evens out the spread of wealth, and theoretically the emergance of middle-classes in developing countries will create enough new wealth to offset the losses initially felt in wealthier countries.
If the scientists are influential, then they're important. The fact that their insights are being followed and expanded upon by others makes them important.
And it's a pretty good guess that they're competent too. If they weren't good scientists then why would their colleages be influenced by them?
Redhat's business model is:
That's they're business model and they aren't changing it. They're just focusing their market. In the desktop market, parts 1&2 don't allow them to compete with MS. The features needed to compete don't exist in the OS, and RH doesn't have the man-power to add them in itself.
In the server market, which RH is currently targetting, they can very effectively compete. So RH is expending it's limited resources to help in compete in this market instead.
there's this stereotype that all asian men are desperate people. The entire culture is surrounded about finding/having boyfriends/girlfriends. Every track on every asian CD, especially the ones you find in hong kong have this same theme, and every TV "drama" has the same themes. In both highschool and college, it seems that meeting someone is their ONLY purpose of existence. When you meet up with friends or family relative members that you haven't seen in a long while, 3 sentences within the conversation is ALWAYS whether you have a boyfriend/girlfriend yet.
And that's different from North America how?
Because the customer reps lie, lie, lie. EVERY time I've had a rep try to sell me on the service plan I'm told it covers accidental damage. Contrary to the fine print two pages into the product service plan details.
And since the product service plan _itself_ doesn't need to be signed, just the receipt, the employee often keeps the service plan in their hands while completing the sale. The employee thus shows only the side with the big point form friendly details not the fine print.
It isn't unreasonable that people are deceived into buying the PSP, and it's rather bigoted to claim they're morons.
Hmm,
1. Instead of offering a $20 rebate and redeeming 50% of those, why not just offere a straight at the counter $10 discount. Pisses off less people
2. I bought a TV-tuner card a couple years ago because it advertised a $50 rebate. Buy it, open it up to read to rebate information inside the box only to find out the rebate expired months before I purchased it. Was I pissed off? Damn straight.
Because clusters are cheaper, per raw unit power.
But if the supercomputer is more efficient per raw unit of power, then the price per unit doesn't matter.
I work for living with HPC, buth with clusters and with large SMP machines. The cluster is nice, but there are some things than can _only_ be run a large SMP machine or are much, much faster on a SMP.
Right and no. If engineers/scientists need to wait for results, then two things happen:
1. Run a cluster and wait two weeks. Twiddle thumbs in mean time.
2. Run on supercomputer. Start processing next day.
For a lot of companies, it is justifiable to spend more money to acheive #2.
Not too mention that a lot of large programs that can't be parallelized have high memory requirements. If your program needs 50GB of memory and can't be parallelized to run across a cluster then you have no choice but run on a supercomputer, since no cheap system can handle 50GB.
Clusters don't scale, huh?
No, they don't (not really anyway). Google isn't a HPC cluster by the way.
Take the big 1000+ CPU clusters build by labs and universities. They aren't running the same application parallelized across 1000 nodes. What happens is UserA gets 16 CPUs, UserB gets 32, maybe 10 for UserC.
Most calculations don't scale well beyond a certain number of processors, and often adding more CPUs will _slow_ down performance. Depending on the problem, a large supercomputer can scale better than a cluster.
Some applications are parallelized to run on 500, 1000, more nodes but those are very rare and very hard to write.
Hmmm...
1. Cray is definitely pro-linux. It's what their XD1 runs. Though not their bigger computers.
2. There are some problems for which that a cluster can not even come close to achieving the performance of a supercomputer. For a lot of problems yes, for some maybe if you spend a fortune on fancy interconnects, and for some no.
3. If you're commercially building clusters let me know company it is. I'm in the market for a 128CPU cluster and I want to know who not to buy from.
Are you being funny or serious?
There's an entire branch of parallel application which are labeled "embarrassingly parallel". This description simply means that such programs are trivially parallelized and achieve as close to linear as possible when scaled across many nodes. This is because of the low inter-node communications.
For "embarrassingly parallel" applications, a cluster is a really good tool. For programs that parallelize as nicely a nice big vector or smp will do nicely. Some code will run better on small 20CPU SMP machine than on a 1000 node cluster.
Let's experiment.
1) Company A raises $100M through debt-financing, talking a 20year loan at a 5% interest compounded annually. Over the 20 years has an average pre-debt-payment profit of $8M. Given yearly debt-payments of $6M, this leaves $2M/year for the company founder, who then has a net worth at the end of 20 years ~$40M.
2) Company B raises $100M through an IPO, with the founder retaining 20% of shares. Given the same profit, the company is free to give out say $6M in dividends (take off some from taxes). The founder then receives $1.2M for a total of $24M at the end of 20 years.
So it appears you're right then and the IPO way isn't as good....right? But wait, the founder still has stocks initially worth $20M which now puts him out ahead. But, then consider that given a standard P/E ratio of 30 the market capitalization for the company is ~$240M, making the founders stake worth $48M, giving him a total worth of $72M.
Of course there's a whole host of other things that affect things one way or the other. Like personal income taxes. Founder A is paying tax on $40M, while Founder B pays on only $24M, since stocks aren't taxed until you sell of them.
Before there was Intel x86 (comptabile) and a number of niche processors, and now there's still Intel and a number of niche processors. The submitter's closing statement seems a tad alarmist.
We still have Itanium, two Sparc variants, a number of Power variants, Transmeta, Opteron, and whole bunch of other niche processors, most of which probably have more market share than alpha.
Someone needs to take an economics class.
1. Yes, in fact you are debt free. No HAH's about it.
2. (almost) No stockholder is going to purposely devalue stock they own just because you don't obey their whims. They are in this to make money afterall, not to megalomaniacally micromanage comparies.
3. Once that $50M is raised initially the stock could go down to $0, and it wouldn't change a thing.