Message boards :
Number crunching :
Thoughts on Credits
Message board moderation
Author | Message |
---|---|
Ace Casino Send message Joined: 5 Feb 03 Posts: 285 Credit: 29,750,804 RAC: 15 |
When I mentioned a few weeks ago that the credit system does not give you any idea how much you have actually accomplished. I stand by that statement. There where many that told me whetstone + cobblestone - drystone = bluestone. Who cares! The fact is wheather it takes you 5,000 seconds or 50,000 seconds to complete a work unit you get the "same" credit. I was under the impression with your "fancy" formula that either you would get more credit for being faster or more for taking longer, or at least there would be a difference (NOT). You can see the number of "results" (or work units completed) by clicking on your s/n. That's the number we should be using "results"! Cause we are all getting the same credit no matter how long it takes so why not use "results" or WU's completed. From> A little Cruncher :) http://www.boincsynergy.com/images/stats/comb-1279.jpg [img][/img] |
N/A Send message Joined: 18 May 01 Posts: 3718 Credit: 93,649 RAC: 0 |
If you can find the records of all the WUs sent, all the hosts that were used (with benchmarks and times), I have no problem sifting through it to tell you (and everyone else for that matter) whether you're right or wrong. The view from behind this keyboard is that credits are nice, but they are not the only thing to BOINC (Except for BRUP because I intend to cash in by cobblecents for a few belches of my own, but that's far down the road). We now return you to your regularly scheduled thread (assuming the scheduler is online...) |
Ace Casino Send message Joined: 5 Feb 03 Posts: 285 Credit: 29,750,804 RAC: 15 |
Click on your s/n, for instance "NA"...than click on computers. It shows your results...you have 3. Its easy and that is the number we should be using. [img][/img] |
N/A Send message Joined: 18 May 01 Posts: 3718 Credit: 93,649 RAC: 0 |
But you're not taking into account this machine (also mine) that had been spewing back garbage results for a day and change (until I found out about it, that is). If we go by the number WUs alone, I'd have 1000+ on a bubkis abacus. Personally, I'd rather see both used. But like I said, until someone gets the records/census of all of SETI/BOINC's WUs and does the statistical voodoo (that who do so well?), this debate will go on ad infinitum or ad nauseum, whichever comes second. |
Digger Send message Joined: 4 Dec 99 Posts: 614 Credit: 21,053 RAC: 0 |
> > > Click on your s/n, for instance "NA"...than click on computers. It shows your > results...you have 3. Its easy and that is the number we should be using. > Randy, the number you're refering to pertains only to the number of work units currently in your results history. Your history is wiped from time to time, so that number is not accurate. Mine shows 58 right now, but i've done more than twice that. I can pretty much guarantee that NA has crunched more than 3 work units. ;) Dig |
Magenta Send message Joined: 10 May 01 Posts: 305 Credit: 6,813 RAC: 0 |
> But like I said, until someone gets the records/census of all of > SETI/BOINC's WUs and does the statistical voodoo (that who do so > well?), this debate will go on ad infinitum or ad nauseum, whichever comes > second. I've been watching the comments in this thread and pondering over the question of credits. One easy way, of course, is just to provide one credit for each successfully completed WU. However, that could put off people who are running slower machines (who would take longer to achieve their credit, assuming similarly sized WUs). So, if one wanted to take that into account one could weight the time taken to crunch as an inverse of some combination of - say for argument's sake (as I'm not a hardware expert) - CPU power, cache and RAM. In this scenario, it would mean that a slower machine got more credit for an equivalently sized work unit than would a faster machine (but they would get the same amount of credit against chronological, as opposed to CPU, time). In the weighted scenario, more machines would still equal more credits. |
N/A Send message Joined: 18 May 01 Posts: 3718 Credit: 93,649 RAC: 0 |
I must've done at least 100WUs by now (not including fubared WUs). That's why I keep insisting for all of the WUs' histories. Only then will we be able to say (to within a 3σ, of course) who's talking out of their hat and who's pulling it from out of their butts. Present company excluded and deluded, of course. |
Paul D. Buck Send message Joined: 19 Jul 00 Posts: 3898 Credit: 1,158,042 RAC: 0 |
If ou go into the FAQ I wrote you will see the whole debate. As far as I know that part of the FAQ is longer than any of the other parts. The bottome line is that if you want to count WU you can. I use BONIC View and take the logs and load them into a database that you can see part of the output if you look for "average processing time" (use search). But quite simply, WU counts only work within a project. Saving the entire list of WU processed would make the DB far to big to be practical on the hardware available at this time, so every project "prunes" the DB back to only the most recent data. The real science results are moved to off-line databases where speed is not an issue. If you look at the WU numbers being issued you can see how many have been created by BOINC, we are already well into millions. |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
> But quite simply, WU counts only work within a project. What about the short and long work units sent out by LHC? Most take under an hour, and then they start sending out the ones that take 6 times longer.... Presumably, we'll get that someday with SETI as well, either through Astropulse or with new telescopes.... |
Paul D. Buck Send message Joined: 19 Jul 00 Posts: 3898 Credit: 1,158,042 RAC: 0 |
> > > But quite simply, WU counts only work within a project. > > What about the short and long work units sent out by LHC? Most take under an > hour, and then they start sending out the ones that take 6 times longer.... Well, if you look at my average processing time page you will see that I did about 1300 work units this last week ... well, one of them was a CPDN, so, that is a hold over ... :) I was just saying that if you ignore length issues, counting WU within a project can be done. But if you want to do that, you have to do it on your own. And I will point out also, that with PHC@Home even the longest potential runs can fail part way through the analysis. Sometimes within seconds. As a matter of fact I did 50+ Wu this last week with 0 CPU time ... is is the old bug? Just dying due to instability before the first checkpoint? :) |
Ace Casino Send message Joined: 5 Feb 03 Posts: 285 Credit: 29,750,804 RAC: 15 |
Keep the credit system just as it is. I have no problem with that. Just add a WU count for the "fun" of it. You don't need to keep the history, just the count. I'm just in it for the fun and would like to know when my credits are in the 10's of thousands or maybe even 100's of thousands (one day) how many WU's that i've done. I BET YOU WILL WONDER THE SAME THING IF YOU DON'T ALREADY......mmmmmm....how many have I done? http://www.boincsynergy.com/images/stats/comb-1279.jpg |
mikey Send message Joined: 17 Dec 99 Posts: 4215 Credit: 3,474,603 RAC: 0 |
> I've been watching the comments in this thread and pondering over the question > of credits. One easy way, of course, is just to provide one credit for each > successfully completed WU. > > However, that could put off people who are running slower machines (who would > take longer to achieve their credit, assuming similarly sized WUs). So, if > one wanted to take that into account one could weight the time taken to crunch > as an inverse of some combination of - say for argument's sake (as I'm not a > hardware expert) - CPU power, cache and RAM. In this scenario, it would mean > that a slower machine got more credit for an equivalently sized work unit than > would a faster machine (but they would get the same amount of credit against > chronological, as opposed to CPU, time). > > In the weighted scenario, more machines would still equal more credits. > But Boinc is not JUST a Seti project, it is a Climate Prediction project, a Pirates Project, an Einstein Project, etc, etc.. To have one workunit equal one credit would mean that people doing different projects cannot compare their results with others doing the same or even different projects. And THAT is the whole point of the credit system under boinc! Not it is true that Berkeley could provide a system to count credits strictly within Seti and whoever could do the same for Climate Prediction but that is each projects decision to make, and the people at Seti have decided not to do that. After all they are the authors of Boinc! |
Bruno G. Olsen & ESEA @ greenholt Send message Joined: 15 May 99 Posts: 875 Credit: 4,386,984 RAC: 0 |
actually I've found another good reason not to use wu count as measurement (or whatever you'd call it). When sah get's to the point where different length wu's are issued, the amount of complaints in the message boards about "why am I getting much fewer of the short wu's" as well as questionable tricks on "how to get short wu's every time" would be flooding the boards if wu count was used. Not to mention serious attempts to cheat for shorter wu's. |
Digger Send message Joined: 4 Dec 99 Posts: 614 Credit: 21,053 RAC: 0 |
Bruno... congrats on User of the Day!! Woo hoo! (At Einstein that is) Dig |
trlauer Send message Joined: 6 May 04 Posts: 106 Credit: 1,021,816 RAC: 0 |
I don't monitor or care about credits for BOINC. I know that a lot of people do care A LOT about it though. There is one project that I run that I do monitor. The client is from United Devices and the two projects I run on it are the LIGandFIT cancer research project and the other is the Rosetta (which is similar to protein folding, I think). What I like about United Devices' "credit" system is that they use total CPU time as one indicator and total points. As you can imagine, I don't care about the total points. What I do care about though is that in the little more than a year that I've been running United Devices' LIGandFIt cancer research project and in the last six months the Rosetta project, I've accumulated total CPU time of 7 years 247 days 22 hours 21 minutes and 23 seconds. I do have this running on between 15 and 25 machines (I don't remember the exact count and some of them aren't on all the time. I'd say it's closer to 18 PCs that are powered on 24x7.) This is a number that tells me that if a scientist were doing the research instead of a PC, that it would have taken the scientist 7 years 247 days 22 hours 21 minutes and 23 seconds to do the same amount of research that my PCs have done in about 13 months. That's the kind of "credits" that I like to see, and it actually has more meaning (to me, anways). Torrey Lauer Modern Travel Services moderntrave DOT net rainbowskytravel DOT com Milwaukee, WI |
Saenger Send message Joined: 3 Apr 99 Posts: 2452 Credit: 33,281 RAC: 0 |
> This is a number that tells me that if a scientist were doing the research > instead of a PC, that it would have taken the scientist 7 years 247 days 22 > hours 21 minutes and 23 seconds to do the same amount of research that my PCs > have done in about 13 months. > That's the kind of "credits" that I like to see, and it actually has more > meaning (to me, anways). The other possible meaning: You spent 7.7 years crunching on old giveaways, whereas the same amount could have been done in less than 1 year on a decent puter. (I didn't look at your puters, and they are probably faster, so nothing personal for you). Raw CPU-time says nothing about work done. Work done is operations done. Flops is the only real good measure. But it's far too much to do this, so estimates have to be done, based on the crunching power and CPU-time. And that's just what credits are doing. Gruesse vom Saenger For questions about Boinc look in the BOINC-Wiki |
trlauer Send message Joined: 6 May 04 Posts: 106 Credit: 1,021,816 RAC: 0 |
> The other possible meaning: > You spent 7.7 years crunching on old giveaways, whereas the same amount could > have been done in less than 1 year on a decent puter. (I didn't look at your > puters, and they are probably faster, so nothing personal for you). Actually, UD (United Devices) sets overall performance of your PC compared to their standard PC. For example: Type Comparison PC This PC Overall 100 168 Processor 100 (P4 1.5Ghz) 155 (P4 2.6Ghz) Memory 100 (384MB) 300 (1280MB) Storage 100 (5GB) 196 (9.77GB) Network 100 (Intel Pro100 S Management adapter) 73 (Netgear GA311 Gigabit adapter) You know my CPU time for UD as stated in my post above. This places me at number 5,975. As far as points generated, I'm at 1,610,958 which places me at 3,182. For results returned, I'm at 3,932 which places me at 12,517. It also tells me I've averaged 23.91472 points per hour, averages 3,977.67407 points per calendar year, averages 409.70448 points per result, and average 9.70864 results per calendar day. I have one PC that is faster than the one I use daily. It's a P4 3Ghz with 1GB of memory. It returns almost twice as many results as my P4 2.6Ghz with 1.25GB of memory. Torrey Lauer Modern Travel Services moderntravel DOT net rainbowskytravel DOT com Milwaukee, WI |
FalconFly Send message Joined: 5 Oct 99 Posts: 394 Credit: 18,053,892 RAC: 0 |
> Cause we are all getting the same credit no matter how long it takes so why not use > "results" or WU's completed. From> A little Cruncher :) That is a common thought error made by new BOINC participants. While a slow machine roughly gets the same Credit as a fast machine per given WorkUnit, the fast one will naturally complete many more WorkUnits per day, thus receiving more Credits per any given time compared to the slow machine. Now add up a slow, singe CPU System against someone using a whole Farm or Cluster, you'll see why Credits 'do' reflect total System performance just fine. But personally, I also find the "Cobblestone" somewhat abstract, having the number of WorkUnits as well as total CPU time listed alongside would indeed be nice to have. |
Bruno G. Olsen & ESEA @ greenholt Send message Joined: 15 May 99 Posts: 875 Credit: 4,386,984 RAC: 0 |
|
MikeSW17 Send message Joined: 3 Apr 99 Posts: 1603 Credit: 2,700,523 RAC: 0 |
> actually I've found another good reason not to use wu count as measurement (or > whatever you'd call it). When sah get's to the point where different length > wu's are issued, the amount of complaints in the message boards about "why am > I getting much fewer of the short wu's" as well as questionable tricks on "how > to get short wu's every time" would be flooding the boards if wu count was > used. Not to mention serious attempts to cheat for shorter wu's. > > Actually sah already does send different length WUs - the amount of processing for any WU will vary acording to the data it contains. More generally though the time taken to process a WU is affected by the Angle Range - I can never remember if high or low angels are faster or slower. The difference between WUs from high-low ranges can be around 10-20% processing time. Thats another reason for credits over WU count - you're credited for what you've actually done. I'm sure that this is an actual cheat in seti classic - there was at least one program going round that looked at each WU in the cache and dumped those with non-optimal ARs. |
©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.