Message boards :
Number crunching :
Credits? Confusing...
Message board moderation
Author | Message |
---|---|
Sirex Send message Joined: 3 Apr 99 Posts: 2 Credit: 10,234 RAC: 0 |
New to bonic seti, and just got a tonne of credits. Some are lower and some are higher than the number in the 2nd last column in the results page... just wondering how they work, because some say like * | 85.57 | 62.72 for example |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
Hi Every WU is sent out to 4 hosts, when they are completed they become claimed credits. When all 4 results returned the highest and lowest will be deleted and a middle of the others will be granted. For better understanding read Paul D Bucks Faqs. greetz Mike With each crime and every kindness we birth our future. |
Sirex Send message Joined: 3 Apr 99 Posts: 2 Credit: 10,234 RAC: 0 |
see whats happening now... just a little confusing the fact that some of my credits are in the 60+'s and i got like 30 odd for them thanks :D |
p Send message Joined: 7 Dec 04 Posts: 106 Credit: 15,334 RAC: 0 |
you get the average of the 4 other pcs that do the same WU not your value (sometimes ur claimed credit is higher than granted credit) but sometimes it is lower and you get granted more credits!!! <BR>AMD XP3200+ <img src="http://boinc.mundayweb.com/seti2/stats.php?userID=2327&trans=off"><img src="http://petrus.homeftp.org/bws/counter_big.php?id=7828479"> <a href="http://homepage.ntlworld.com/paulandrew.odell/">MY SITE!</a> |
B-Roy Send message Joined: 4 May 03 Posts: 220 Credit: 260,955 RAC: 1 |
the thing is that linux machines claim much lower credit. if you crunch the wu with let's say 3 other linux-users, your granted credit will turn out to be much lower than what you claimed. but in the average, this effect smooths out. |
Paul D. Buck Send message Joined: 19 Jul 00 Posts: 3898 Credit: 1,158,042 RAC: 0 |
> For better understanding read Paul D Bucks Faqs. And I am sure that I point out that it is an evil deamon in the BOINC Code that is responsible for these problems ... ") |
Neil Walker Send message Joined: 23 May 99 Posts: 288 Credit: 18,101,056 RAC: 0 |
> the thing is that linux machines claim much lower credit. And Macs and Intel H-T machines running Windows .... ;) In general, my Linux machines claim about twice as much credit as a P4 H-T running Windows XP. But, then, I have taken the trouble to compile my own Boinc clients. ;) Be lucky Neil |
mikey Send message Joined: 17 Dec 99 Posts: 4215 Credit: 3,474,603 RAC: 0 |
> > the thing is that linux machines claim much lower credit. > > And Macs and Intel H-T machines running Windows .... ;) > > In general, my Linux machines claim about twice as much credit as a P4 H-T > running Windows XP. > > But, then, I have taken the trouble to compile my own Boinc clients. ;) > I am NOT trying to suggest ANYTHING here! But if you or I can compile our own clients, that then ask for whatever amount of credit, isn't that akin to cheating? Am I missing something here?! Probably!!!!! I already know we can compile our own client, but did not know that those could then ask for more credits than usual. Is the Berkeley software so unteaked that you are now actually requesting the "proper" amount of credit for a machine like yours? |
Neil Walker Send message Joined: 23 May 99 Posts: 288 Credit: 18,101,056 RAC: 0 |
> Is the Berkeley software so unteaked that you are now actually requesting the > "proper" amount of credit for a machine like yours? Well, it's two things really. First, the Microsoft compiler "cheats" by over-optimising. It effectively eliminates some of the loops in the benchmark. The Berkeley developers are trying to find a solution to prevent that and have gone at least some way to doing that in the latest versions. Second, GCC, the linux compiler, is a very complex piece of software due to the fact that it has to support so many different platforms. With it's default settings, it generally produces code that will run but not necessarily the best code. By tuning it to the hardware involved, a more realistic benchmark can be achieved. Now, just to complicate matters further, for some reason, Intel H-T processors seem to return very poor results for the benchmarks which do not reflect their actual performance on WUs. Maybe someone better versed in Intel technology can explain that. Be lucky Neil |
mikey Send message Joined: 17 Dec 99 Posts: 4215 Credit: 3,474,603 RAC: 0 |
> Second, GCC, the linux compiler, is a very complex piece of software due to > the fact that it has to support so many different platforms. With it's default > settings, it generally produces code that will run but not necessarily the > best code. By tuning it to the hardware involved, a more realistic benchmark > can be achieved. > So does EACH user have to tune it to their system to get the optimal settings? |
Frymaster Send message Joined: 3 Apr 99 Posts: 1 Credit: 64,600 RAC: 0 |
I wondered why my brother's AMD 2500 got a higher integer score than my Intel 3GHz... (well, why it got a higher score than twice my Intel, but you know what I mean...) EDIT: The way it was explained to me about why different systems claim different amounts: The amount claimed is related to the number of clock-cycles it takes the processor to process the unit. Older processors can be inefficient, so take more cycles. Hence an old Pentium might crunch a unit in a day or so, and claim 50 credits. A more modern CPU might only take a couple of hours, but only claim 30 credits. What this will tend to mean is "inefficient" processors are penalised ("It might have been difficult for *you* but this guy found it east") and "efficient" ones benefit. |
Neil Walker Send message Joined: 23 May 99 Posts: 288 Credit: 18,101,056 RAC: 0 |
> So does EACH user have to tune it to their system to get the optimal > settings? Some of the more experienced or adventurous Linux users will use a distribution like Gentoo where everything can be compiled from source code (as I do) and each will tend to have their own preferences for settings but the vast majority of Linux users rely on distributed binaries and don't give optimisation a thought. Be lucky Neil |
Toby Send message Joined: 26 Oct 00 Posts: 1005 Credit: 6,366,949 RAC: 0 |
One of the users here has gone to the trouble of compiling the linux core client for various hardware platforms and has made them available to everyone here. These clients tend to claim approximately what normal windows boxes claim. As for HT: The problem is that a HT CPU does not give you the same performance as 2 CPUs - especially with heavy crunching like the benchmark does. When the benchmark runs, it sees 2 CPUs so it runs 2 benchmark threads. However they interfere with each other since they are running on the same CPU so the benchmark score is much lower than it should be. Seti@home however does benefit a bit from HT so the work units finish in less time than they should, compared to the benchmark scores and therefore the core client claims less credit. Credit is claimed not based on the number of operations performed but on the time it took to do the work combined with the benchmark scores. > Older processors can be inefficient, so take more cycles Actually this is not true... at least for Intel. The P3 core was and still is way more efficient than the P4 core. Late P3s outperformed early P4s. The P4s only pulled ahead because they cranked up the GHz. The Pentium-M chips used in laptops are based on the P3 core and they blow any P4 of equal GHz out of the water. FYI :) A member of The Knights Who Say NI! For rankings, history graphs and more, check out: My BOINC stats site |
Paul D. Buck Send message Joined: 19 Jul 00 Posts: 3898 Credit: 1,158,042 RAC: 0 |
> I am NOT trying to suggest ANYTHING here! But if you or I can compile our own > clients, that then ask for whatever amount of credit, isn't that akin to > cheating? Am I missing something here?! Probably!!!!! > I already know we can compile our own client, but did not know that those > could then ask for more credits than usual. Is the Berkeley software so > unteaked that you are now actually requesting the "proper" amount of credit > for a machine like yours? One of the other aspects that I don't see addressed here is the fact that you can claim all you want, but the process of granting credit (which varys some from project to project) basically throws out the high and low claims, averages the rest and grants a median value. Other projects may simply grant the lowest of the claims, the middle claim, etc. ... However, none of the projects grants the high score ... So, yes you could "doctor" the BOINC Manager code and make insane claims, but it does not do you any good. Unless, and until, every one uses the compiled version that does make those silly claims. Since a goodly amount of us are just in it for the science and the fun ... well, I (for one) would not use a BOINC Manager that out of whack ... |
Neil Walker Send message Joined: 23 May 99 Posts: 288 Credit: 18,101,056 RAC: 0 |
> So, yes you could "doctor" the BOINC Manager code and make insane claims, I haven't seen anyone advocating such a practice here (and, again, it's the daemon - not the manager ;) ) > but it does not do you any good. It does everyone some good to level the playing field a little. Take the following hypothetical case. Two Linux users with Berkeley-compiled clients and a Windows XP user make the following claims: Linux1 23.87 Linux2 22.91 Windows 39.42 Each user will be granted 23.87. However, suppose Linux2 was using an optimised client: Linux1 23.87 Linux2 38.63 Windows 39.42 In this case, each will be granted 38.63 > Since a goodly amount of us are just in it for the science and the fun ... That's just a teeny bit patronising, Paul. :( For many, the competitive element is the fun. If it were not for those people with outrageous seti farms in their spare bedrooms, sheds or wherever and the network admins who install BOINC on 200+ machines at work, there would be precious little science being done. > well, I (for one) would not use a BOINC Manager that out of whack ... DAEMON, Paul, ;) All we are suggesting is a way to bring the Linux (and other) versions up to their full potential and more into line with the (majority) Windows users. Be lucky Neil |
mikey Send message Joined: 17 Dec 99 Posts: 4215 Credit: 3,474,603 RAC: 0 |
> One of the other aspects that I don't see addressed here is the fact that you > can claim all you want, but the process of granting credit (which varys > some from project to project) basically throws out the high and low claims, > averages the rest and grants a median value. > > So, yes you could "doctor" the BOINC Manager code and make insane claims, but > it does not do you any good. Unless, and until, every one uses the compiled > version that does make those silly claims. Since a goodly amount of us are > just in it for the science and the fun ... well, I (for one) would not use a > BOINC Manager that out of whack ... > Yes that is true but if you could crank your version up a bit and still stay within the "mathematical limits" of the 3 or 4 units returned you could crank up the credits awarded to everyone that crunches the same units as you do. And since that is highly variable, you would be the one benifitting, not the others, because their interaction with you is on a rare unit by unit basis, while yours is on EVERY unit. AGAIN I am NOT suggesting this is the case!!! I was curious and the answers have been enlightening. Basically I have learned that while the client works as is, it could stand a bit of "tweaking" for all platforms. SOME users have done this on their own, most have not. I think that Berkeley should download the different versions and perhaps come out with a better client for all to use. Some of the users seem to have put a lot of thought into the process. Berkeley could perhaps learn from them. |
Paul D. Buck Send message Joined: 19 Jul 00 Posts: 3898 Credit: 1,158,042 RAC: 0 |
> I haven't seen anyone advocating such a practice here (and, again, it's the > daemon - not the manager ;) ) Yeah, I am still thinking "old" where there pretty much is just the one glob of stuff. I am trying (I know very trying). > That's just a teeny bit patronising, Paul. :( Hmm, wasn't intended as such. But it is true. I actually think most of us are into it for the science above all. But the vocal minority that rants on about credit wears even the saints down ... I was saying with no malice or overtones what I meant. Just to annoy my spouse, "Autism is always saying exactly what you mean." ... > For many, the competitive element is the fun. If it were not for those people > with outrageous seti farms in their spare bedrooms, sheds or wherever and the > network admins who install BOINC on 200+ machines at work, there would be > precious little science being done. Well, I will just point out that I too have a small farm, and for the most part they run 24/7 ... and as soon as I save up enough and the latest G5 comes out, well, my count will go up by one. > > well, I (for one) would not use a BOINC Manager that out of whack ... Yeah, yeah, see above... > DAEMON, Paul, ;) All we are suggesting is a way to bring the Linux (and other) > versions up to their full potential and more into line with the (majority) > Windows users. Actually the fix is to bring the Windows machines down to the rest of the world. Again, I am not trying to create a war where none existed before ... |
Neil Walker Send message Joined: 23 May 99 Posts: 288 Credit: 18,101,056 RAC: 0 |
> Actually the fix is to bring the Windows machines down to the rest of the > world. Not quite because, apart from the Windows client over-performing on the benchmarks, the Linux (and, I think, other) clients as compiled and distributed by Berkeley are under-performing. The devs are working on the first issue but not the second at the moment. Of course, the real fix is a more realistic benchmark - such as giving each host a standard WU to crunch. ;) Be lucky Neil |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13736 Credit: 208,696,464 RAC: 304 |
> The Pentium-M chips used in laptops are based on the P3 core Very, very loosely based on the P3. > and they blow any P4 of equal GHz out of the water. In some applications, but not others. Grant Darwin NT |
mikey Send message Joined: 17 Dec 99 Posts: 4215 Credit: 3,474,603 RAC: 0 |
> Of course, the real fix is a more realistic benchmark - such as giving each > host a standard WU to crunch. ;) > Now that idea I LIKE! Maybe a small unit that takes about an hour on say a 1.6ghz machine. Something to benchmark your system against AND ensure that your machine is performing "PROPER" calculations. After the one unit is processed more units could be sent depending on your cache settings, 1 day, 2 days, etc.. Of course on HT machines you would get 2 of the units, they would just be identical. |
©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.