A story of success
20 Mar 2008 22:53:17 UTC
Topic 193582
(moderation:
Today it´s 1/2 year that we´re crunching S5R3.
During that time the crunching speed has increased by 198 +/- 6 [GFlops/d], as linear fitted, and the number of "active hosts within the last 7 days"/7 has increased by 76,7 +/- 3,5[hosts/d]. Within the last 2 weeks these figures are even much more impressive: The crunching speed was changiging by 575[GFlops/d] and the number of active hosts 785 +/- 43[hosts/d]. That´s a realy impressive growing up. We hope for further groth!
At the 11th march we had the first time a crunching speed of 100[TFlops]! There are only a few computers in the world of this capability.
If one takes the ratio of the "Total Users" as seen for E@H on http://www.boincsynergy.com/stats/ and the "active hosts within the last 7 days"/7 form the E@H Server Status page, one get something like the relative number of active hosts [%]. These figure had a minimum of 30% at Sept. 07, and is now back to 39%, but still far away from the 56% we had in oct. 06. These doese show to me, the importance of the impressive reliability of the farm of servers feeding the project.
On the basis of the crunching activity within the last week, we will finish the S5R3 project within about 120 days from now on. This comes out from the "estimated work still remaining" on the server staus page, which is dropping by 2,2[d/d] within the last week.
M.W.
A story of success
)
Some of the credit goes to the wonders of Moore's Law. Those Intel Core 2's and beyond are really speedy and more power efficient. Hardware advances just keep coming!!! Now one pc box is really 4, 8, 16, .... computers as far as BOINC projects go. Now my V8 Mac Pro purchased in Aug 2007 is already a slow poke compared to new machines.
The E@H project may also be getting new users because of the stability of the project and its overall good quality. So a hearty good job to the E@H team!
I think that Einstein is
)
I think that Einstein is getting some new users because of the willingness of the project to dedicate programming time to getting the most out of donated systems, which doesn't go unappreciated or unnoticed apparently. I pretty much stopped all of my PCs from sharing time on other projects simply because this one is run so well and the science behind it is so compelling.
I agree that Einstein is
)
I agree that Einstein is probably the most reliable project and that the Project leaders are involved and keep us informed.
But I am a bit of a sceptical about this being the only reason. I believe a lot of the recent interest is due to the power users application. e.g. a pent M 1.8GHz running power user application completes a unit quicker than a core2 2.4GHz running default application.
RE: I agree that Einstein
)
I'm skeptical about the credit reduction that I know has to be percolating by now... If it is done like in the past, the SETI optimized apps will become the higher paying of the two...
RE: I'm skeptical about the
)
They already are for Macs IME—both PPC & Intel—by a margin that I estimate to be about one-third. OTOH the stock SETI@home app for Mac/PPC (based on my experience with the Beta project) yields less than E@h, in a roughly similar proportion.
RE: RE: I'm skeptical
)
A quick mental maths check on my Pent M, new accounts about two weeks ago, which does Einstein, Seti and Seti Beta adjusted for resource share says that;
Einstein on Power app would get ~800/day
Seti using Optimised app would get ~600/day
and Seti Beta on standard app would get ~400/day.
Hi! I'm running CPDN and
)
Hi!
I'm running CPDN and Einstein@Home with 50/50 share on my MacMini and both yield about the same credits/hour. The Mac E@H app is comparable to the Linux power user app app (which is now also the official Linux app). I guess it's not (yet) necessary to think about a credit reduction in terms of "fairness" to other projects.
CU
Bikeman
RE: On the basis of the
)
Historically, a lot of people start shutting down their BOINC computers from May through August. So, during the summer, we'll see the active hosts figure go way down. That would also imply that it will still take more than 120 days to finish the current run.
Hi! Yup, here's a graph of
)
Hi!
Yup, here's a graph of active hosts over the last year, the "summer effect" is clearly visible:
Once the Windows App gets to the level of the current Linux version (auto switching for SSE optimized app), I think E@H will be able to offset the loss of hosts in summer by the increased performance per host, so I guess a progress of 0.33% per day should be sustainable on average.
With 58 % to go in S5R3, this would mean another 175 days or so, or less than 1 year for the whole S5R3 run. Not so bad!
CU
Bikeman
5 days ago I set the
)
5 days ago I set the following resource share(setting, not %) for the following projects on two quads and 3 dual cores.
Malaria Control 50
Einstein 25 (using 4.38 for linux)
ABC 25
Rosetta 25
Basically, after some time the cpu time and credit if equal should be the same for Einstein, ABC, and Rosetta. Malaria control should be twice the others.
Here's the credit/day according to Boincstats for the last 4 days (23 mar, 22 mar, 21 mar, and 20 mar):
Einstein
3794, 3322, 4271, 2609
ABC
2368, 2712, 2489, 1423
Rosetta
1162, 1042, 1338, 1246
Malaria
2357, 2214, 2104, 1857
Prior to this run, I ran WCG and Einstein alone, both at 25 and here's the credit/day for those days (mar 18, 17, 16, and 15)(I've excluded data on the 19th as a transition day)
Einstein
6640, 6638, 5689, 6165
WCG
1933, 1922, 2431, 1972
I'd like to think the higher credit/day for Einstein is a function of it's previously filled "pending threshold" while the other projects were filling the threshold(Rosetta has no pending as credit is granted nearly instantly), and will see more as time passes. I'm letting the "work scheduler" fill the cache as it will and overtime it should represent equal amounts of cpu time. Connection interval 0.01, Additional days 0.33.
this certainly isn't accurate enough to make any accurate conclusion as to the amounts, but does seem to indicate Einstein pays more/time than other projects. Note: Rosetta pays about 7% less than the old "Benchmark system" would have (this is due to averaging, and many older attached hosts using the pre 5.9.3 or there about boinc version compiled with MSVC 2003 as opposed to the MSVC 2005 which yeilds higher benchmark scores). ABC is being run on 64b linux OS with 64b linux app so ABC score is influenced by the more efficient app. I have my other single core machines NOT attached to these projects, but attached to other projects, so other projects will show credit, but it's not from the two quads and 3 dual cores and vice versa.
I'll just let it run (atleast until I hit my credit goals/project, or Boincsimap has more work) and see how this changes. Boincstats credit/day graph is below and linked so it should change day/day when updated.
Let's just say that over a year ago, I gave up working/recording data to show real percentages. Basically, the cross project parity idea is so badly bent that no comparison can be made using "boinc cobblestones", unless each user has exactly the same resource share, cpu run seconds etc. Someone at 400K cobblestones may have actually done LESS work than someone at 100K, so Boinc Credit is worthless, unreliable, and not worth anything if trying to make comparisons. It's way to late for anyone to correct it, unless Boinc finds a way to enforce equality and steps up to do, then to start a whole new set of stats.