Message boards :
Number crunching :
Here's how to Get More Credit
Message board moderation
Author | Message |
---|---|
Bill & Patsy Send message Joined: 6 Apr 01 Posts: 141 Credit: 508,875 RAC: 0 |
[Warning: the following posting includes a few tongue-in-cheek loaded words, like "cheated" and "deserve", because I was having fun touching on some topics that from time to time have been controversial. Flame back at me if you like, but I'd prefer that you enjoy this posting as it is intended to be: playfull. Apart from the "tone" of the presentation, however, I do believe that the theory itself is sound, and suggest that you give it a try. Enjoy!] I have several machines that constantly get cheated out of the credits that they have honestly earned. Some machines claim large credits per work unit, others claim very little credit. When the quorum is determined and credits are awarded, machines in the quorum that claim very little credit consistently cause my high-credit machines to get much less credit than the true credit that they have earned and deserve. I suspect that many others have similarly been cheated out of the credits that they deserve. Here's the solution, and it will help people with low-credit machines too to get higher credit awards as well. In other words, EVERYONE will benefit. SETI will benefit too since the overall average Cobblestones awarded per work unit will go up for the entire program. Solution: If your machine claims high credits, keep a very small cache so that your results are reported very early and your claimed credits are definitely part of the quorum that determines the credits. Conversely, if your machine claims high credits, keep a very large cache so that your results are reported very late (but within the 14 day limit, of course) so that your claimed credits are definitely NOT part of the quorum that determines the credits. That way, the higher credits on average will be the ones that determine the points that are awarded, and the lower credits will not be part of the determining-quorum and will therefore not pull the award level down. (Note that this only works as long as SETI continues to waste computational power by sending out 4 work units to get a 3-unit quorum. If they ever revert to the more efficient 3-on-3 configuration, this strategy will then only work when a work unit has to be re-sent, which would not happen very often.) I have been experimenting with this over the last month or so, and it definitely works. Of course, it doesn't work on very single work unit, but on average, more of my work units are getting higher overall credit awards now than they used to. That applies to both the "high" credit machines (that now have small caches) and to the "low" credit machines (that now have very large caches). If everyone did this (which won't happen, of course), then low-credit machines would never form part of the determining quorum; only high-credit machines would be the ones determining the credit. Life would be much better, we would all be much happier, and we would all live much longer. (Well, maybe, maybe not.) Seriously, though, the theory is sound. Try it. Happy crunching, and may you receive the higher credits that you deserve!! --Bill |
MikeSW17 Send message Joined: 3 Apr 99 Posts: 1603 Credit: 2,700,523 RAC: 0 |
Damn! Caught with pants down. LOL |
Tigher Send message Joined: 18 Mar 04 Posts: 1547 Credit: 760,577 RAC: 0 |
|
RichaG Send message Joined: 20 May 99 Posts: 1690 Credit: 19,287,294 RAC: 36 |
Maybe your high credit claiming machine is asking for more than it deserves so you would then be cheating. It's all in how you look at it. Red Bull Air Racing Gas price by zip at Seti |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
Yep, exactly that's the way it works :P ;) Aloha, Uli |
Skip Da Shu Send message Joined: 28 Jun 04 Posts: 233 Credit: 431,047 RAC: 0 |
[Warning: the following posting includes a few tongue-in-cheek loaded words, like "cheated" and "deserve", because I was having fun touching on some topics that from time to time have been controversial. Flame back at me if you like, but I'd prefer that you enjoy this posting as it is intended to be: playfull. Apart from the "tone" of the presentation, however, I do believe that the theory itself is sound, and suggest that you give it a try. Enjoy! OK OK, but here's another procedure that will net you more credit than all the cache tweaking you can do... besides, the new scheduler in cc v4.4x negates about 90% of that teaking. Go over to Home Depot and buy a new lawn mower on one of those 12 months no payments, no interest deals. Sell your existing existing lawn mower and hit NewEgg with the cash for a cheap mobo with video and lan built in and a cpu (best bang for the $ is probably AMD xp 2xxx for around $75). Head to the used computer junk store (or Goodwill computer store) for an old 250w or less PSU. Find an old plastic basket and some wire ties and build a basket cruncher ;-) Now you've got a new lawn mower (I'd go with a rider) and another number cruncher and you're wife is barely even aware of what happened. LOL Oh yea, I've got a couple used riders for sale if anyone is interested ;-) - da shu @ HeliOS, "A child's exposure to technology should never be predicated on an ability to afford it." |
rattelschneck Send message Joined: 14 Apr 01 Posts: 435 Credit: 842,179 RAC: 0 |
Skip, does the word "density" ring some bells? ;) regards |
Tigher Send message Joined: 18 Mar 04 Posts: 1547 Credit: 760,577 RAC: 0 |
|
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
Classic just plain classic....vertical farm in a vegetable stack!. Great just great! Can it do the dishes after dinner? Probably not, but you could use Skip's plan to go to Home Depot on their 12 months no interest credit and buy a new dishwasher, sell the old one and put that money in at NewEgg for more processors and motherboards.... |
Tigher Send message Joined: 18 Mar 04 Posts: 1547 Credit: 760,577 RAC: 0 |
Classic just plain classic....vertical farm in a vegetable stack!. Great just great! Can it do the dishes after dinner? LOL! Not sure exactly what new egg is but I get the drift.......the secret in his solution was the wife never finds out. She might spot a new dishwasher! |
Prognatus Send message Joined: 6 Jul 99 Posts: 1600 Credit: 391,546 RAC: 0 |
Remember guys that this is an international project (and fora), so not everyone knows about NewEgg, Home Depot and special US deals. :) |
The Pirate Send message Joined: 14 Apr 00 Posts: 191 Credit: 4,929,008 RAC: 0 |
|
Heffed Send message Joined: 19 Mar 02 Posts: 1856 Credit: 40,736 RAC: 0 |
Hmmmm.... Don't I remember you making a similar post some months ago saying that you get credit faster by hanging on to your WUs as long as possible? My same answer still applies to this slightly different theory. It's all dependent on your sample point. The overall outcome is the same, but your observations are skewed by where you take the sample from. Getting your high credit machine in the quorom doesn't increase your chances of getting more credit, as the quorom rules still apply. The median result defines the credit granted. Whether you are first in, or third, the outcome is the same. With the amount of users participating in the project, everyone with high credit machines turning in work quicker wouldn't have a noticeable impact. No matter how quick you report, it's still the luck of the draw as to who you get paired with. |
Bill & Patsy Send message Joined: 6 Apr 01 Posts: 141 Credit: 508,875 RAC: 0 |
Hmmmm.... Don't I remember you making a similar post some months ago saying that you get credit faster by hanging on to your WUs as long as possible? I'm not going to try to explain it to you, Heffed. But if you work on it some more, you'll figure it out. Start by considering that the sample involved here is four, not three, and consider the effect if the first three to be returned are higher, and the last to be returned is lower. This won't happen, as I previously said, with every work unit, but it will happen more often than otherwise if you see to it that YOUR result is ALWAYS either IN the quorum or OUT of the quorum, depending respectively upon whether you're claiming high or low credits. It might help you to consider a model in which all results throughout the SETI world are returned that way, then back up to the real world, but skewed by your own choices concerning timing of your returned results. It's a statistical thing, Heffed, which I'm sure you can figure out. (And note that in my posting I pointed out that this wouldn't work if SETI returned to the three work unit paradigm, which is what your attempted rebuttle is based upon.) Concerning my prior posting about getting credits faster, you are correct in your recollection. So, the trade-off for high-credit-claiming machines is: do you want your points awarded faster when your results are returned, or do you want more points when they are finally awarded. I have informally tested both approaches, and they are both valid, theoretically and in practice. Apart from that, it's nice to hear from you again! Hope things are going well for you! Best regards, --Bill |
Ned Slider Send message Joined: 12 Oct 01 Posts: 668 Credit: 4,375,315 RAC: 0 |
/me thinks we need a "post pics of your farm" thread, vegetable stacks or otherwise :p Ned *** My Guide to Compiling Optimised BOINC and SETI Clients *** *** Download Optimised BOINC and SETI Clients for Linux Here *** |
jimmyhua Send message Joined: 16 Apr 05 Posts: 97 Credit: 369,588 RAC: 0 |
Hey, that's an interesting theory. It's about distributions, assuming there are equal numbers of PC's claiming high medium and low credit.... Let's think about this. What are the chances that tweaking your cache will make a difference? Let's do some probability counting: There are 4 guys, they will either claim high credit, low credit, or medium credit. Let's enumerate all the possibilities. LLLL LLLM LLLH LLML LLMM LLMH LLHL LLHM LLHH LMLL LMLM LMLH LMML LMMM LMMH LMHL LMHM LMHH LHLL LHLM LHLH LHML LHMM LHMH LHHL LHHM LHHH MLLL MLLM MLLH MLML MLMM MLMH MLHL MLHM MLHH MMLL MMLM MMLH MMML MMMM MMMH MMHL MMHM MMHH MHLL MHLM MHLH MHML MHMM MHMH MHHL MHHM MHHH HLLL HLLM HLLH HLML HLMM HLMH HLHL HLHM HLHH HMLL HMLM HMLH HMML HMMM HMMH HMHL HMHM HMHH HHLL HHLM HHLH HHML HHMM HHMH HHHL HHHM HHHH 81 different possibilities. Let's start with a low credit machine. If you had a low credit machine that claimed credit last so you won't be part of the quorum. Would it help in claiming more credit? Well, in the case of LLLL, hell no as seti will take the median which is low credit. In the case of LLLM, still no as the median credit will still be low. But for LMHH. YES it would turn the tide, or LMMH, no it won't. Anyways, enumerate every possibility and start counting. LLLL LLLM LLLH - no, no, no LLML LLMM LLMH - no, yes, yes LLHL LLHM LLHH - no, yes, yes LMLL LMLM LMLH - no, yes, yes LMML LMMM LMMH - yes, no, no LMHL LMHM LMHH - yes, no, yes LHLL LHLM LHLH - no, yes, yes LHML LHMM LHMH - yes, no, yes LHHL LHHM LHHH - yes, yes, no MLLL MLLM MLLH - no, yes, yes MLML MLMM MLMH - yes, no, no MLHL MLHM MLHH - yes, no, yes MMLL MMLM MMLH - yes, no, no MMML MMMM MMMH - no, *no, *no MMHL MMHM MMHH - no, *no, *no MHLL MHLM MHLH - yes, no, yes MHML MHMM MHMH - no, *no, *no MHHL MHHM MHHH - yes, *no, *no HLLL HLLM HLLH - no, yes, no HLML HLMM HLMH - yes, no, yes HLHL HLHM HLHH - yes, yes, no HMLL HMLM HMLH - yes, no, yes HMML HMMM HMMH - no, *no, *no HMHL HMHM HMHH - yes, *no, *no HHLL HHLM HHLH - yes, yes, no HHML HHMM HHMH - yes, *no, *no HHHL HHHM HHHH - no, *no, *no So, out of 81 possibilities, 32-> yes, turning in your results last helps 16-> *no, (this is possible in the grand scheme of things, but you'll never see it as your computer is supposed to always underclaim credit) 33-> no So, there's a 32/65 chance that the outcome is improved thanks to ensuring that your claimed credit is not part of the quorum. We can redo the count for making sure that a high credit claiming machine is always part of the quorum. However, that would be cheating! It's another exercise in counting, and I think it'll probably be close to 50% chance that it will help the outcome. Now, if *everybody* did this. It actually wouldn't make much difference, as you'd still get paired up with the possibilities laid out above. So it would still only improve the credit awarded, about half the time. The other half of the time, you would either have some serious bad luck, or get the credit you *ahem* "deserve." Jimmy |
BotLobsta Send message Joined: 22 May 05 Posts: 4 Credit: 51,547 RAC: 0 |
"jimmyhua" wrote: Let's start with a low credit machine. Just one thing about your tables. I cant figure out what the *'s mean, but I noticed that those appear whenever there is no low-credit claiming machine. Unless I am missing something, this makes no sense because you are assuming that your own machine is low-credit. I believe someone once defined compromise as a solution that neither side is happy with. |
jimmyhua Send message Joined: 16 Apr 05 Posts: 97 Credit: 369,588 RAC: 0 |
Just one thing about your tables. I cant figure out what the *'s mean, but I noticed that those appear whenever there is no low-credit claiming machine. Unless I am missing something, this makes no sense because you are assuming that your own machine is low-credit. I think you have the right idea. If you're just looking at your own machine and what affect the credit will have, then it makes no sense to look at what happens when seti server at berkeley issues the WU to a HHMM combo, you're not part of it. If *everybody* started to do as Bill suggests, and you wanted to know what the overall effect would be. You would have to start counting these too. Hence, the *'s. Jimmy |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
Classic just plain classic....vertical farm in a vegetable stack!. Great just great! Can it do the dishes after dinner? Home Depot is a "big box" discount home improvement store. They usually have some credit deal where you get 12 months of no interest financing. They sell building materials, appliances, lawnmowers, etc. NewEgg sells computer parts -- they're at NewEgg.com and have been mentioned many times here in the fora. ... and I think Skip's plan is sheer genius: a way to get a new lawnmower and some surreptitious cash. Besides, I'm sure he loves his wife and can probably convince her that she deserves a new dishwasher. |
mikey Send message Joined: 17 Dec 99 Posts: 4215 Credit: 3,474,603 RAC: 0 |
I'm not going to try to explain it to you, Heffed. But if you work on it some more, you'll figure it out. Start by considering that the sample involved here is four, not three, and consider the effect if the first three to be returned are higher, and the last to be returned is lower. This won't happen, as I previously said, with every work unit, but it will happen more often than otherwise if you see to it that YOUR result is ALWAYS either IN the quorum or OUT of the quorum, depending respectively upon whether you're claiming high or low credits. It might help you to consider a model in which all results throughout the SETI world are returned that way, then back up to the real world, but skewed by your own choices concerning timing of your returned results. It's a statistical thing, Heffed, which I'm sure you can figure out. (And note that in my posting I pointed out that this wouldn't work if SETI returned to the three work unit paradigm, which is what your attempted rebuttle is based upon.) The fly in your ointment is multiple machines, those of us with "farms" CANNOT do what you are suggesting. I can see your point but i think it is waaaay too complicated and as indicated could be negated by changes in the next version of the schedulat, which would in turn mean you having to try and find a new way to "beat" the system. Since there is only 1 cache level setting, and those of us with farms would have both high and low credit machines your system is useless! If you only have one machine, what the heck is the point if only 50% or so of the time it works? That means that a little less than 50% of the time is does not work. Pure random dumb luck could negate those odds. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.