Is our participation useful?
21 Jun 2008 12:43:25 UTC
Topic 193730
(moderation:
I've just pulled the statistics for top 500 participants into Excel and found that 2 participants on the top do more work then 500 others and so on.
I think now that our participation is unuseful, because ATLAS when working on this project when it is idle of other tasks do much more then thousands of us.
Moreover it has more power efficiency per watt that most of us have.
So, I beg Bernd and others to open our eyes on this dark side of computing. Is our power still useful for the project or we can step by step move to the others or even turn off our rigs just for the economy?
Is our participation useful?
)
I don't think ATLAS is running E@H full time, it has other things to do. But I find often Steffen Grunewald for Merlin/Morgane as a wingman and my Opteron 1210 competes bravely with his Opteron 185.
Tullio
RE: I don't think ATLAS is
)
Yes, but I mean not a core per core comparison. I mean that all the cores ATLAS consist of are doing more sense to the project in its idle time, than 500 most powerful users of the project including myself.
It now looks like this is the competition between us, users. But these 2 grids on the top of the participants list is out of order.
The picture will become clear if the ATLAS would be included in the list of participants as an individual with starting point from 0 (zero) credits.
Then we all will see how much it works. And it will be then obvious after some time, whether all of us try to compete or not.
P.S. You can compare your Opteron with one core in ATLAS, but you can not compare you box, or say, Bikeman's farm with all those 5000 cores in ATLAS working simultaneously.
RE: RE: I don't think
)
Why would you try to compare an individual PC with a supercomputer? That's like trying to compare a nuclear bomb with a firecracker.
Seti Classic Final Total: 11446 WU.
RE: ... 2 participants on
)
In another way of looking at the data, the top two have recently been contributing somewhat less than a third of the daily output. While we "little folks" are individually tiny (or, in your case, small) compared to them, collectively we are contributing significantly more than them.
{speculation}It is even possible that our participation weighs a little in funding decisions, so possibly we even get a little "virtual credit" for them{/speculation}
Your point on power efficiency has some real merit. At SETI's rollout, people thought of computation for it as free use of "wasted cycles", but for a long, long time now the dominant processor/OS combinations in use in fact consume far more power when running BOINC than when idling. Just to provide one example, the moderately overclocked and overvolted Q6600 system, with more than average peripherals, on which I am typing this comment dropped from about 223 watts with BOINC running to 142 watts on stopping BOINC just now. I get over 4000 Einstein cobblestones/day from those 81 watts, but it certainly is not free.
I think folks who are burning extra power on older machines by running BOINC on them, especially if they leave them powered on far more hours per year than they otherwise would, really should think about the balance of considerations. I, myself, run a 930 MHz Coppermine 24/7. It used to have to run, as it was my printer host, but an E6600 in the same room picked up that task. On my own advice, I really should consider stopping BOINC on it, and powering it up only when required for the scanner it hosts or other purposes. Or, as I am actively contemplating, I could replace it with a modern build, probably using a Penryn-generation processor, but neither overvolted nor overclocked. I think I could match the Coppermine host's total system power consumption in a machine with well over ten times the Einstein output.
I know my computers do only a
)
I know my computers do only a really small part
of the whole work, but I joined to Einstein@Home
because it is one of the most interesting projects
that offers the possibility of participation. I belive
that one of my computers will discover the secret
of the gravitation. :-)
RE: I know my computers do
)
Dream on. . .
We all know that it'll be one of my computers! ;)
RE: RE: I know my
)
Baloney!!
I have it on good authority the honor of the first gravity wave will go to a slug (and one of mine at that)! :-D
Alinator
RE: I've just pulled the
)
G'Day Stranger7777,
I believe that we are greatly needed for the amount of data that is needed to be processed, now and in the future.
Even if ATLAS had 10,000 or 20,000 cores that is only equal to from one or five universities participation.
There are over 100,000 active users, so even if they only had one computer each, they would still have five to ten times the processing power of ATLAS.
It maybe powerful but it can't do the job on it's own, we are needed.
>> Keep on smiling as it makes others wonder what you have been up to.
I think our participation is
)
I think our participation is still useful and needed.
If the Atlas Cluster delivers about 30 Teraflops
and the project goes on with nearly 160 Teraflops,
the smaller contributions make the bigger part
of the computing power.
Another question is, if our contribution is energy-saving
or increasing the power dissipation of the project.
I tried to find an answer for myself. I'm using a System
with an Intel Core 2-Duo-Processor E8200, running at 2666 MHz.
My graphics card ist consuming up to 45 Watt, the whole system,
when idle, uses 100 Watt. While Einstein is running, the power
goes up to 110 or 112 Watt. The LDC uses 26 Watt extra.
If i give my idle-time to Einstein@home, those 10 to 12 Watt
of additional power-consumption give a computing power of 5600 Mips
(Whetstone). I think this gives an excellent ratio of
Mips / Watt. In a specialized computercluster they nearly don't need
graphics cards and LCDs, but the active cooling is power-consuming too.
I estimate, Atlas yields a five to ten times lower ratio of Mips/Watt.
On the other hand, if i let my system run only for Einstein@home,
its far less effective than a node in a well-managed cluster.
And if one is using a gaming-system, overclocked and with one or two
fat graphics cards, he should better let the professionals do the job
and switch off his his electric heating.
But i must admit: Sometimes, i switch off only my LCD end let the system work for Einstein ...
RE: I think our
)
That is a bit unfair to ATLAS, I guess. The 30 TFlops figure for Atlas is derived from the Linpack benchmark, the 160 TFlops for the E@H network is derived from the formula
(daily credits for the entire project)/ 100 credits * 1 GFlops
which has "historic" reasons more than anything else.
If you compare the RAC from PCs with Xeon cores comparable to those that ATLAS is made of, I think it's more realistic to estimate that ATLAS would score about (roughly) 7 Mio credits a day so E@H as a whole is roughly twice as powerful as ATLAS, and roughly equivalent to a supercomputer with 60 TFlops Linpack performance. Still extremely impressive!