Slashdot first reported that Jim Gray (a Microsoft researcher) explains the relative cost of supercomputing/clustering/gridding for various types of task.
Cost Economics of Grid Computing
Submitted by Will Sutton 2003-07-08 General Development 10 Comments
Hmm, mustn’t be prejudiced by the fact that it’s a Microsoft guy! Very
interesting article, which does well in explaining the concepts in terms
that anyone can understand, using nice simple back of the envelope
maths. I never got my head around all this grid stuff; this certainly helped to clarify it.
Other people have commented on it before, but Mr. Gray didn’t talk about security concerns. People find it difficult to trust computers outside of their control with their data.
Web based email is a real success in terms of establishing trust…but the @home distributed computing projects are different. My impression of the @home projects is that most people donate or “sell” their time because the applications offer the promise of community-wide profit. Maybe OS news could run a pole on the reasons people run x@home.
Online gaming is the primary commercial venture that I can see that has a chance of making real money. Music and video sharing (distributed ripping and encoding) have of course proven very popular with consumers, but it sure isn’t popular with industry (Apple is successfully selling digital music, but it’s a centralized process again due to the whole trust issue). It is interesting that Mr. Gray didn’t mention distributed audio/video ripping and encoding. If only the music industry could see distributed music encoding and storage from Microsoft’s viewpoint, they were getting advertising, years of computation, and massive amounts of redundant storage for free! Instead of remaining calm and taking advantage of their new tremendously powerful distributed computer, they panicked, destroyed it, and lost consumer trust/goodwill.
At any rate, I think that the number of applications suited to distributed computation is lower than Mr. Gray thinks. The technological infrastructure for distributed computing makes it more and more feasible and attractive for more applications, but changing people’s perspective and habits is more difficult.
As soon as I saw that the article was written by a Microsoft employee, I had to discredit everything.
Sorry to reply to my own post, but I reread it =-)
“It is interesting that Mr. Gray didn’t mention distributed
audio/video ripping and encoding. If only the music industry could see distributed music encoding and storage from Microsoft’s viewpoint, they were getting advertising, years of computation, and massive amounts of redundant storage for free!”
The above didn’t come out right…I was trying to poke a little fun while making a point, but it came out nonsensical. Here is another try:
It is interesting that Mr. Gray didn’t mention distributed
audio ripping and encoding ala Napster when he mentioned renderfarms. If only the music industry was as excited about distributed computing as Microsoft and the general public are. From Microsoft’s viewpoint, they were getting advertising, years of computation, and massive amounts of redundant storage for free!
There, I think that is better. It really is too bad that the music industry felt they could not trust their consumers to buy the music they were getting for free =-) I admit that it does sound kind of illogical; but of the people I know that have shared music online, all of them continued to buy about as many CDs as before.
I hope this wasn’t offtopic…I was just trying to point out how big an issue trust is when computing.
Do not let the fact that he works for microsoft discredit him. He’s a renowned expert and his point is clear, concise and well made.
Compared to a lot of non-microsoft writers who’s work you can access through osnews, he is one of the few that really knows his shit.
He makes his points but they are based on common sense intuition and not fact; and therefore is fundamentally flawed. And in his conclusion he failed to make the sale. What was he really trying to say? It’s all nonsense and a waste of my time to read and write this reply.
I disagee with “He makes his points but they are based on common sense intuition and not fact”. All sorts of semi-quantitative arguements can see like ‘nothing more’ than common sense. Agreeing with common sense is usually a good thing. Besides, he does have quantitative arguements. He compares how the costs of network traffic and CPU utilization affect distributed computing. In scaling arguements, there is usually no need to quantify the prefactors with high precision. He notes that communications costs are not dropping as quickly as computation (or storage) costs & correctly (IMO) uses that fact to argue that grid computing is limited (in many cases) by the cost of networking. So computations that are distributed over the internet only make sense the tasks are very cpu intensive. It seems like a very good sort of ‘back of the envelope’ calcuation, which I tend to trust more than sophisticated calculations that can only be done on a computer 😉
Obviously the author is NOT a business economist. Mixing up fixed and variable costs is never a good thing, escpecially when you try to calculate TCO. The second mistake was mixing up different cost elements.
No valid conclusion can be made basing on the authors assumptions.
I think this is a lightweight article that summerises Gray’s understanding. And Gray knows his stuff (Turing Award winner etc) in this field.
I am not surprised that he mentioned Java instead of .Net. I am however surprised that no-one here tried to twist that 😉