It also confirms Sek's other graph which suggests that recently the WU have got that bit longer to crunch

geopsychic

It is an interesting graph. I think that without further background information you may draw erroneous conclusions. Mr. Autumns has, I'm sure, the correct explanation for the initial peak. There were far fewer computers on the project at that time and with workunits being in the 100+ hours range, it makes sense. As they got better at paring down the size of the work units, it makes sense that the graph would flatten out.
There are a number of factors in play which could affect the outcome of the average time:
1. The relative number of long workunits vs. short ones (Fa@H vs.GC)
2. The ratio of speedy computers to slower ones
3. The completion of projects and the addition of projects over time.
4. The targeting of various projects to particular kinds of machines.

For instance, if you look at the stats for RCTC Grid and Xtreme Systems. Xtreme Systems is probably running all high end fast machines whereas RCTC Grid is probably running stock P4's. Xtreme Systems turns in many more results on less run time. If you knew the ratio of WU were identical for each, you would know Xtreme Systems had faster machines. However, if they were only specifying GC units and RCTC Grid were getting a mix of all the projects, the results returned figures could also reflect this disparity.

Also, with the advent of GC and its lower requirements, I was able to put several machines into use which would otherwise have been idle. These are all lower end PII and PIII's with neither the memory nor speed to attempt any other current project on BOINC. When I look at the results from my slowest machine under workunit status I see that that it is about 6 times slower than the fastest result, i.e. 12 hours vs. 2 hours. This addition of lower end machines, if large enough, could have had the effect of raising the average daily time.

Gradually as people upgrade their systems and retire older machines, the average daily time per result should decrease if the size of the WU's remains on average about the same as it is currently.

It would be interesting to know the percentage of machines at various benchmark strata. If these figures were known, an efficieny coefficient could be calculated. If the mix of the sizes of the WU's would remain constant, this would give an indication of the efficiency of the grid as a whole.

I am sure that someone with a statistical background could come up with all sorts of different ways to measure the work and would be able to give an explanation of what the implications for those figures might be.

You have provided food for more thought. Thanks

Cheers

Sgt.Joe

Minnesota Crunchers

Hey
Did a small bit of deriving from the graph ( thank you for the graph! ), fumbled some coordinates, and did a snakes butt regression analysis. My hand to screen skills being what they are, here goes a brief synop...
Robust marks show between 7/05, and present. Probably when the ud grid took a bad dump, wcg picked up crunchers... The lows, delineated by x's, are wcg downtimes; note the 12/06 midmonth fall when boinc was down. Any increases following these downs are just fair weather sunshine. but the rise,in late /06 is indicative of 1) More projects 2) More cpu's online 3) More networked cpu's online 4)The advent of significant amounts of dual core cpu's online, expanding the range of results in the domain of a narrower list of projects... It just might be that another marker for cpu time, maybe as a factor being project particular, can be considered for scoring or cutting pieces of pie!

Just to cloud / clarify the discussion here are the CPU time, results and points graphs.

There is a strong weekday vrs weekend modulation - no real surprise there except possibly the amplitude.