The 64000 Cobblestone Questions
12 Aug 2008 21:32:34 UTC
Topic 193850
(moderation:
From the official Berkeley documentation:
1 Cobblestone = 1.728e12 Combined Operations
Looking at each science run individually:
1.) How many operations were there in the the median S3Rx task?
2.) In the median S4Rx task?
3.) In the median S5R1 through R3 task?
4.) If I had an old S5R4 task which took 345.60e12 operations to complete few years ago, and a new S5R4 task today which takes 345.60e12 operations to complete, what would they be worth in Cobblestones?
5.) If I used the performance metrics of the median host few years ago to set the basis for the credit computation, and then use the median host today to set the basis, will the answers to Question 4 be the same?
6.) Assume that the basis for a project was set reasonably close to the definition of the Cobblestone at some point in the past. If one has a host which was accumulating credit at a rate of 10 Cobblestones per hour with a given application, what should its credit rate be now for a new application compiled to the same standards as the first one?
7.) Given a task with the same amount of work as in Question 4 and using the same host as Question 6:
Assume that older application did not take advantage of all it's mathematical computational features. You now run the task through it with an application which does.
What effect would this have on the value of the task in Cobblestones?
8.) What would be the effect on the rate for the host in Question 7?
9.) Suppose the answer to Question 5 is no. Then how can one say the Cobblestone is a fixed and constant quantity for measuring the contribution a host provides to a project over time?
10.) Can you see any possible problems with the CPP strategy recently proposed by Berkeley?
Alinator
The 64000 Cobblestone Questions
)
Typo Corrections:
1.) The 'e' in the mathematical notation does not represent the base of the natural logarithm. It represents 10 to the power of whatever follows it. I should have use 'E' for that, just like my trusty HP calculator does.
2.) In Question 4: "...old S5R4 task..." should be "...old S4Rx task...".
Alinator
You do realize that only us
)
You do realize that only us 'old farts' will remember the reference to "The $64,000 question", right? ;)
I wish I could answer the questions, but I'd barely know where to start trying. :-/
RE: You do realize that
)
LOL...
That was the idea... I crossed over into the old fart, dinosaur category not too many moons back, but my memory still works fine! ;-)
Alinator
RE: RE: You do realize
)
I was going to say something, but I forgot what it was. ;)
Seti Classic Final Total: 11446 WU.
That's tough! Before even
)
That's tough!
Before even beginning to answer the question, we would have to agree on what a "combined operation" really is.
For example, if an app uses an SSEx instruction that multiplies 4 pairs of floating point numbers and then adds a vector of 4 other floating point numbers, would that be 1 operation, 4, or maybe 8? And how are we supposed to count the operations after all?
CU
Bikeman
RE: That's tough! Before
)
Simple, Combined Operations (COP) is merely the sum of all the integer and floating point operations it takes to complete a computational task put to a computer. The figure I gave for the value of the Cobblestone in Combined Operations was derived directly from the specifications of the BOINC reference computer, so I suppose one could say it's integral to and comes directly from the givens.
Your example of introducing SSE to the question is precisely what is wrong here. You cannot set the basis using anything other than straight integer and FPU math, because that is how the Cobblestone is defined. IOW, those X operations 'packed' into one SIMD instruction still count as X operations for the purpose of scoring.
The fact you can speed up execution of a given application using the advanced computational features does not mean that algorithmically you have simplified the 'problem', or the amount of work contained in a given task is reduced because one machine can execute it faster (more efficiently) than another. This is the conceptual mistake of treating power and work as the same thing.
Your final question to my question goes right to the heart of the issue! :-)
The real problem is that I don't know of any commercial, commodity processor which has the capability built in to account for the amount of work it does as it's been defined here. That means we are left with having to calculate the base property, work, from it's derived quantity, power, and that's where all the fun and games begin as it stands. ;-)
Alinator