| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: BOINC Agent Support
Thread: Got a problem |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 22
|
|
| Author |
|
|
Movieman
Veteran Cruncher Joined: Sep 9, 2006 Post Count: 1042 Status: Offline |
Just a small interjection from a non-tech, but if the RAM cant keep up with the CPU, you are actually running longer than the CPU could do a job in. The benchmark, flawed as it is, only tests the CPU! No, the limitation isn't in that way but as it effects front side buss speed. The speed at which the data crosses the system itself. That's the limitation that we get with dual socket systems that use ECC REG memory. There are also 4 'lanes" for lack of a better word that are shared. I saw the effect when I ran a dual clover but with only one cpu installed. The single at 2667 at that time was getting 14,000 a day but the dual at 3000 at that time was getting just barely 25K. There is cpu speed and overall speed of the system. The memory effects only the latter.  [Edit 2 times, last edit by Movieman at Aug 20, 2007 7:46:12 AM] |
||
|
|
Movieman
Veteran Cruncher Joined: Sep 9, 2006 Post Count: 1042 Status: Offline |
Still seeing the same issue here.
----------------------------------------The overall is that if I get matched with a slow machine I wind up getting his claimed points and when I get rarely matched with a machine thats even close to mine in speed only then is an average of the 2 machines taken. Now since I'm using a stock unaltered Boinc client I don't understand how the WCG servers could see me as overclaiming. I'm using the suggested Boinc client from their site: 5.10.13 http://boinc.berkeley.edu/download.php Many of my teammates have seen this also and it's frustrating. We're bring the best equipment to the project and it appears like we're being penalized for doing so. I know that there was no malicious intent in this but is it fair or reasonable to award points to anyone using a random other machine as an equalizer? Assume for a minute that machine A is a Dell P4-2000 with 512mb ram. The old side of the average home machine. It's running 40-45 processes, hasn't been defragged or had it's caches cleared in a year and runs at maybe 65% of it's true capabilities. Machine B is a dedicated cruncher. Thats all it does and is highly tuned and maintained. There is just no comparision between the two and to do so is just plain not right. Also, whose machine do you think is giving you the truer benchmark of what it is doing? The one that JUST does WCG or the one that might be running or not running any of several apps when boinc decides to bench every 5 days? Then there is the factor of the many different Boinc agents involved that all bench differently. As long as that is allowed to take place you'll never see "fair" claims because there is no set benchmark but many. I hope you can hear my frustration in this. I thought on this and tried to come up with an answer. The only thing I can think of is this: 1) send the WU to machines running the same version of boinc OR make a statement that only X version of BOINC will be allowed. 2) send the WU to machines in the same speed range and break it down to 500mhz increments: 1000-1500,1501-2000,etc up to 4500mhz That isn't a "fix all" but it will make things a heck of a lot more realistic than they are now. |
||
|
|
Diana G.
Master Cruncher Joined: Apr 6, 2005 Post Count: 3003 Status: Offline Project Badges: 
|
Movieman Dave, I always have the same point scale as you. I always have more points than the other guy. The range is only within 20 points, so it doesn't eat me up inside. I actually look at the positive: look how our higher score pulls up the other person's score :-) It must give them an excited and happy feeling to want to keep crunching and I bet they say "Thank you, whoever you are for being my partner on this work unit!"
----------------------------------------Brings a smile to my face everytime I see a greater granted credit than a claimed credit for the other guys  I just know that a completed WU is a completed WU no matter what the points we are awarded add up to.  Diana G.  |
||
|
|
Movieman
Veteran Cruncher Joined: Sep 9, 2006 Post Count: 1042 Status: Offline |
Movieman Dave, I always have the same point scale as you. I always have more points than the other guy. The range is only within 20 points, so it doesn't eat me up inside. I actually look at the positive: look how our higher score pulls up the other person's score :-) It must give them an excited and happy feeling to want to keep crunching and I bet they say "Thank you, whoever you are for being my partner on this work unit!" Brings a smile to my face everytime I see a greater granted credit than a claimed credit for the other guys I just know that a completed WU is a completed WU no matter what the points we are awarded add up to. Diana G. Hi Diana; It doesn't "eat me up" but it does bother me. I beleive in "fair is fair" and the real issue is in what is being seen as "normal". Also for the most part I'm not pulling them up, they are getting what they claimed and pulling me down. If it was an average I could live with that but it isn't. It's like reading a statement that the WCG servers don't beleive what I'm sending but they do beleive the other. That is what bothers me on a personal level. I'm doing it nice and clean with the best equipment money can buy on the most stable windows OS that exists and still running into an issue. That is the frustrating part. One of my teammates mentioned that WCG was using a Dell P4-3000 as the norm. The machine I have in question here probably has 12 times that machines computational power in total. Thanks for commenting. I know your motives were good. |
||
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3716 Status: Offline Project Badges: 
|
Movieman, I have a medium-slow machine (P4 HT 2.4 GHz) and according to the Results Status page I often pull a faster machine up! Less often it is the opposite. And most of the time both scores are different by a few points. This not to mean (and even less demonstrate) that your case is wrong but that adjustments really happen both ways and that it is not always the faster machine which is penalized.
----------------------------------------Two rather large differences still visible in my results: - me: 9.06 h 60.4 credits vs 4.99 h 45.5 credits; granted 60.4 - me: 7.19 h 51.4 credits vs 2.40 h 39.8 credits; granted 51.4 Had it been the average systematically, we would have lost a few credits.  Cheers. Jean. |
||
|
|
Movieman
Veteran Cruncher Joined: Sep 9, 2006 Post Count: 1042 Status: Offline |
Movieman, I have a medium-slow machine (P4 HT 2.4 GHz) and according to the Results Status page I often pull a faster machine up! Less often it is the opposite. And most of the time both scores are different by a few points. This not to mean (and even less demonstrate) that your case is wrong but that adjustments really happen both ways and that it is not always the faster machine which is penalized. Two rather large differences still visible in my results: - me: 9.06 h 60.4 credits vs 4.99 h 45.5 credits; granted 60.4 - me: 7.19 h 51.4 credits vs 2.40 h 39.8 credits; granted 51.4 Had it been the average systematically, we would have lost a few credits. Cheers. Jean. Hi Jean, Thanks for the imput.This is what I'm seeing and maybe a picture can say it better than I can. THis is just the last few validated on that machine.  |
||
|
|
olympic
Senior Cruncher Joined: Jun 12, 2005 Post Count: 156 Status: Offline |
Strange indeed. My dedicated crunchers almost always win the "who's closer to their average" decision. Maybe 10% get decided in the other guys favor. I see you run only FAAH, maybe try a different project and see if the machine fares better. Sounds strange but sometimes it works. I know my AMD machines get murdered point-wise in HPF2 but do just fine in the rest.
---------------------------------------- |
||
|
|
Movieman
Veteran Cruncher Joined: Sep 9, 2006 Post Count: 1042 Status: Offline |
Strange indeed. My dedicated crunchers almost always win the "who's closer to their average" decision. Maybe 10% get decided in the other guys favor. I see you run only FAAH, maybe try a different project and see if the machine fares better. Sounds strange but sometimes it works. I know my AMD machines get murdered point-wise in HPF2 but do just fine in the rest. That sort of presents an issue in itself. My interest is in AIDS and cancer research primarly and HPF2 although worthwhile, I see as helping the "esteemed" Dr.Baker at Rosetta and my feelings on him are well known. I can't in all good faith do work for someone I don't respect and in terms of people I don't respect, he's at the top of that list. Now I'm not against doing research on the new Dengue project but I'd much prefer to work on the AIDS project until they start up on the next cancer project. Thanks for the input! |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Movieman, until WCG starts assigning credit on actual ops performed against the cobblestone standard or some other measurement that includes things like total system burden, there is no way around the many variables that enter into the equation. To name but a few
----------------------------------------1. different versions that meet in a quorum 2. stability of benchmark in above 3. the present load during benchmarking of the partners in a quorum (is s/he playing super mario while it happens) 4. the pool mix of OSses particularly on HPF2 I did a dump every 3-4 days from the RS pages into a spreadsheet and found that in a few months time the credit claimed v credit granted gave a negative of 2k BOINC points.... no quad Q6xxx or other screamer in the lot. Resigned to it being like it is.
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 1 times, last edit by Sekerob at Sep 1, 2007 9:56:14 PM] |
||
|
|
Movieman
Veteran Cruncher Joined: Sep 9, 2006 Post Count: 1042 Status: Offline |
Movieman, until WCG starts assigning credit on actual ops performed against the cobblestone standard or some other measurement that includes things like total system burden, there is no way around the many variables that enter into the equation. To name but a few 1. different versions that meet in a quorum 2. stability of benchmark in above 3. the present load during benchmarking of the partners in a quorum (is s/he playing super mario while it happens) 4. the pool mix of OSses particularly on HPF2 I did a dump every 3-4 days from the RS pages into a spreadsheet and found that in a few months time the credit claimed v credit granted gave a negative of 2k BOINC points.... no quad Q6xxx or other screamer in the lot. Resigned to it being like it is. Hi Sek: Thanks for replying. I'm glad to see that you agree that there are many factors that toss off the equation. One question: How long ago was it that you did that spreaddheet? I have a good reason for asking that I'll fill you in on after you answer. Thanks.. PS: One fast little fix that might help this? Maybe a direct link to the BOINC download page from WCG to the current suggested BOINC version? I know WCG suggests 5.8.16 I beleive where BOINC suggests 5.10.13 and they bench very differently.. Maybe that will help get us all closer together???? |
||
|
|
|