| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: Suggestions / Feedback
Thread: Can we get some help in this thread from a CA Please! |
| No member browsing this thread |
|
Thread Status: Locked Total posts in this thread: 70
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Like Movieman, in our team I`m losing good machines to other projects because of the points handicap of Linux when the OS actually is a benefit and crunches WU`s faster upto 50%.
Appreciate your effort Dave but my teamie`s with linux are not convinced you have any clout and are despondently looking for other work !  Please remember a lot of the experts are not of English Nationality plus the fact that this forum is not frequented by all. |
||
|
|
David Autumns
Ace Cruncher UK Joined: Nov 16, 2004 Post Count: 11062 Status: Offline Project Badges: 
|
Hi carl.h I'd have to part agree with this sentiment "are not convinced you have any clout"
----------------------------------------I've been a thorn in IBM's side for long enough for them to know that I won't let it lie. Back in the early days (the threads are all there for anyone to see) I'd like to think I made a difference with a lot of effort on my part. Hey IBM even sent me a WCG T Shirt for my efforts (sad but true) When the email from Big Blue arrived inviting me to become a CA I thought I might be able to make more of a difference on your behalf sadly this hasn't turned out to be the case. Unfortunately for you all us current CA's are the best hope you have and I must quickly make my apologies for my many failings. But at least I'm still in there kicking and shouting on your behalf not all of those 3000+ posts of mine have been made in public and I did have an alta ego of Samuel Clemens for a while  [Edit 1 times, last edit by Former Member at Oct 30, 2006 4:36:09 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hi carl.h,
I realise there is no way a TRUELY fair system could happen but the ludicrous system that really hurts the Linux crunchers is not fair at all or even near it yet no one seem`s to care (that matters). It profers the question is Boinc /WCG in Microsofts pocket ? I realise that Windows in virtually the default OS but a lot of experienced computer people crunch and a big percentage use Linux ! Back in the 1990s, there were a lot of nasty posts about the Windows compiler programmers. The posts said that they were competing with each other by tweaking their compilers to optimize the standard benchmarks much better than they optimized standard applications. Then they would advertise the benchmark scores for their compilers to try to make more sales. (Money talks!) Since BOINC uses standard benchmarks as part of the scoring formula, I think that we are running into inflated scores for applications compiled for Windows. No, I don't KNOW this. But I suspect that this is what we have run into. It would explain the discrepancies between Linux and Windows applications. But this would vary from compiler to compiler. If true, the standard benchmarks are only good for a single version of a compiler. As soon as a project takes advantage of all the different OS versions and competing compilers, we are in the Wild West and anarchy reigns. Solution??  I can't think of one that is easy enough to be practical. Which is why I tend to stay aloof while reading posts arguing about points.Lawrence |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
O.k. I can only talk Boinc as that is my application(?)......
----------------------------------------Let`s take faah WU`s........ Windows averages for these are in the 55-65 Boinc points range which is the Boinc points average for the time worked ( 10+ points an hour on a good machine 3500+) multiply this x7 for WCG points. On Linux on the same box you would see your score through Boinc drop to 25-28 yet see a 50% improvement in the time taken to crunch a WU. Same amount of work done yet half the score. On a super machine such as a E6600 the Boinc points average 20+ per hour due to work going through it.....The time taken is equivalent to the previously mentioned Linux box yet the Linux scores just 7 points per hour. The solution is simple as they did with Rosetta.......A certain type of WU (designed to last X on a computer) is worth Y points. Therefore you are being paid for work done no matter how quick or slow the machine ! Your machine is fast get`s more done you get paid more but there is no variation for OS. This type of sytem is a lot fairer than a time system and would drive upgrades to more powerful machines as they would increase the work rate therefore more points which is fair. [Edit 3 times, last edit by Former Member at Oct 30, 2006 8:03:42 AM] |
||
|
|
David Autumns
Ace Cruncher UK Joined: Nov 16, 2004 Post Count: 11062 Status: Offline Project Badges:
|
the problem with this one carl.h is that the work is described as non-deterministic i.e. random, of no fixed length and there's no way to make it so so unlike a SETI work unit they all have differing length and tend to come through in similar (but not the same) batches.
----------------------------------------Behind the scenes a suggestion was made for a control set of PC's to allocate the given points for a work unit not sure if this is workable but it is sitting with IBM. Morning Lawrence This would be true if ever the WCG moved away from pure X86 to specific CPU optimisations but to date due to time pressures and the envisaged run time of a project these MMX SSE etc options haven't been taken up so a raw benchmark should be just that On the OS front we hear tales of Windows being inefficient and bloated and Linux being swift of foot but are you really saying the same hardware running Linux would complete a WU in half the time every time? Dave |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Behind the scenes a suggestion was made for a control set of PC's to allocate the given points for a work unit not sure if this is workable but it is sitting with IBM. This is exactly what Rosetta do now ! Exact same box ...................I cannot vouch for exact WU`s for reason`s you`ve already stated but the averages are alway`s pretty similar imho. Windows XP SP2 Workunit Name Device Name Status Sent Time Time Due / Return Time CPU Time (hours) Claimed/ Granted BOINC Credit faah0839_ bdb609_ mx1sh9_ dry_ 02 papa Valid 10/21/2006 19:47:23 10/24/2006 06:05:46 5.11 75 / 56 faah0839_ bdb566_ mx1sh9_ 0C papa Valid 10/21/2006 19:47:23 10/25/2006 22:08:16 5.53 81 / 81 faah0839_ bdb601_ mx1sh9_ 03 papa Valid 10/21/2006 19:47:23 10/24/2006 18:06:55 5.11 75 / 75 faah0839_ bdb612_ mx1sh9_ 00 papa Valid 10/21/2006 19:47:23 10/24/2006 18:06:55 5.00 73 / 73 faah0839_ bdb612_ mx1sh9_ dry_ 00 papa Valid 10/21/2006 19:47:23 10/24/2006 06:05:46 5.39 79 / 79 faah0839_ bdb614_ mx1sh9_ dry_ 04 papa Valid 10/21/2006 19:47:23 10/24/2006 18:06:55 5.03 73 / 60 faah0839_ bdb615_ mx1sh9_ dry_ 02 papa Valid 10/21/2006 19:47:23 10/24/2006 06:05:46 4.98 73 / 60 faah0839_ bdb616_ mx1sh9_ 01 papa Valid 10/21/2006 19:47:23 10/24/2006 18:06:55 4.96 72 / 72 faah0839_ bdb620_ mx1sh9_ dry_ 01 papa Valid 10/21/2006 19:47:23 10/24/2006 21:40:22 4.92 72 / 72 faah0839_ bdb620_ mx1sh9_ dry_ 02 papa Valid 10/21/2006 19:47:23 10/24/2006 21:40:22 4.99 73 / 61 faah0838_ bdb384_ mx1sgu_ dry_ 04 papa Valid 10/20/2006 06:05:32 10/23/2006 14:59:37 5.05 74 / 52 faah0838_ bdb380_ mx1sgu_ 0B papa Valid 10/20/2006 06:05:32 10/23/2006 14:59:37 5.41 79 / 78 faah0838_ bdb385_ mx1sgu_ 09 papa Valid 10/20/2006 06:05:32 10/24/2006 06:05:46 5.03 74 / 74 faah0838_ bdb374_ mx1sgu_ dry_ 06 papa Valid 10/20/2006 06:05:32 10/23/2006 14:59:37 5.22 76 / 70 faah0838_ bdb385_ mx1sgu_ dry_ 00 papa Valid 10/20/2006 06:05:32 10/24/2006 06:05:46 5.02 73 / 67 Linux Workunit Name Device Name Status Sent Time Time Due / Return Time CPU Time (hours) Claimed/ Granted BOINC Credit B05057_ 0072_ CTMA4A-36-15-2-c1 Papa-Linux64.fl home Valid 10/22/2006 11:07:54 10/25/2006 17:38:47 1.21 10 / 10 faah0870_ bdb197_ mx2avo_ 05 Papa-Linux64.fl home Valid 10/22/2006 11:07:54 10/25/2006 14:15:27 3.35 27 / 21 faah0870_ bdb204_ mx2avm_ dry_ 08 Papa-Linux64.fl home Valid 10/22/2006 11:07:54 10/25/2006 17:38:47 3.41 27 / 27 faah0870_ bdb203_ mx2avm_ dry_ 06 Papa-Linux64.fl home Valid 10/22/2006 11:07:54 10/25/2006 17:38:47 3.43 28 / 22 faah0870_ bdb205_ mx2avo_ 05 Papa-Linux64.fl home Valid 10/22/2006 11:02:45 10/25/2006 14:15:27 3.59 29 / 33 faah0870_ bdb204_ mx2avo_ 02 Papa-Linux64.fl home Valid 10/22/2006 11:02:45 10/25/2006 14:15:27 3.38 27 / 26 faah0870_ bdb204_ mx2avo_ 01 Papa-Linux64.fl home Valid 10/22/2006 11:02:45 10/22/2006 15:24:45 3.38 27 / 21 faah0870_ bdb201_ mx2avo_ 04 Papa-Linux64.fl home Valid 10/22/2006 11:02:45 10/25/2006 14:15:27 3.50 28 / 28 faah0870_ bdb200_ mx2avo_ 05 Papa-Linux64.fl home Valid 10/22/2006 11:02:45 10/25/2006 14:15:27 3.26 26 / 25 faah0870_ bdb200_ mx2avm_ dry_ 05 Papa-Linux64.fl home Valid 10/22/2006 11:02:45 10/24/2006 22:55:00 3.27 26 / 41 faah0870_ bdb194_ mx2avo_ 00 Papa-Linux64.fl home Valid 10/22/2006 11:02:45 10/25/2006 14:15:27 3.21 26 / 26 faah0870_ bdb194_ mx2avm_ dry_ 00 Papa-Linux64.fl home Valid 10/22/2006 11:02:45 10/25/2006 14:15:27 3.14 25 / 21 faah0870_ bdb188_ mx2avm_ dry_ 01 Papa-Linux64.fl home Valid 10/22/2006 11:02:45 10/25/2006 14:15:27 3.34 27 / 27 faah0870_ bdb199_ mx2avm_ dry_ 01 Papa-Linux64.fl home Valid 10/22/2006 10:57:39 10/22/2006 15:24:45 3.26 26 / 37 This box is a AM2 x2 5000+ [Edit 1 times, last edit by Former Member at Oct 30, 2006 8:20:25 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hi carl.h,
Carefully thought out! Good, but here is the next question - The proposed solution should be seen as fair within the project and should reduce arguments about comparing scores within a single project. However, what do you consider to be the possible results on arguments between people running different projects? Keep in mind, many of the most dedicated crunchers run several different projects. Would the final end state be better, worse, or equally contentious? (That is a real question, not a disguised argument. I don't have any feel for what the answer is, but I am sure that anything we do would involve additional programming time and possible controversies with each new project. So unless the result is a definite improvement, I don't feel like recommending a change.) Lawrence |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Are we talking projects as in the Boinc sphere or the projects in WCG ?
----------------------------------------Let me address a little anomaly namely Boinc and it`s points system. There is no parity in different projects I have proven this in the past, it is getting better but as I have convinced a lot of discerning parties before, the Boinc points are already skewed. Rosetta now keeps to a pretty good average re Boinc points .....A given machine will earn X an hour. [Edit 1 times, last edit by Former Member at Oct 30, 2006 8:43:24 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hi all
Just a quick post as I'm off to a job today, so will be around late tonight. Whilst I myself am not too bothered by points I can see the problems for people who are using Linux and want points. I run Linux and the Bog standard Boinc for Linux. This is through my choice of operating system, I use Simply Mepis 6.0-1, mainly as I like Linux and my main game I play is really fast on this OS. My system is as follows :- Processor : Intel P4 D630 EM64T HT 3.0 Ghz Mobo : ECS 865PE-A7 Memory : 2 Gig DDR400 Graphics : Nvidia Geforce FX 5600 using nvidia-glx Monitor : Emprex LM1904 19" widescreen LCD monitor Sound : Onboard 6 channel ICH15 ( Duplex ) Cooling : Aksa AK-920 Evo 120 Cooler Fan ( Gas Evaporation ) Drives : /dev/hda Maxtor 6E030L0 30 Gig : /dev/hdb Seagate ST313032A 13 Gig CD Writer : /dev/hdc Samsung CD-R/RW SW-W162C DVD Writer : /dev/hdd TssT Corp CD/DVDW SH-W162C Mouse : M$ Intellieye 3 Button Wheel Mouse ( Optical USB ) Keyboard :M$ 105 Key Soft keyboard PS/2 English-uk layout I appologise for the fact it covers everything, but it's what I use on the Linux forums. As you can see it's nothing extraordinary, but with Hyperthreading I do 2 WU's at once and they take approx 5 hours 56 minutes to complete. So you could say that a WU completes every 2 hours 58 minutes. I'll post a couple of examples. faah0874_ bdb491_ mx2avv_ dry_ 03 localhost.local domain Valid 10/27/2006 16:08:15 10/29/2006 15:54:58 5.97 25 / 19 faah0874_ bdb399_ mx2az8_ 0B localhost.local domain Valid 10/27/2006 13:22:35 10/29/2006 10:30:46 5.88 25 / 28 faah0874_ bdb261_ mx2az8_ 02 localhost.local domain Valid 10/27/2006 07:47:28 10/28/2006 19:42:00 6.04 35 / 24 This is a standard Boinc client not optimised and a non overclocked system, as for the points I'll let you lot work them out compared to Windows, as I'm not sure how you convert them to WCG points. carl.h, and movieman, I applaud your crunching and crunching power as individuals and teams and yes I can now get the idea that points are very important to a lot of people, so run time is meaningless for the cost you guys have laid out. I have in all honesty bought the gas cooler and the extra 1 gig so I can crunch and play UT2003 at the same time, where your expense must be massive. Yes I think we have to take serious note of the points issue if we want to keep valuable resources like these teams and individuals. They are now I realise employed and payed by points and as in any workforce pay discrepancies are a very valid point and should be addressed as quick as possible. One other thing guys can you let me know how my crunching is spec wise for time and points ( I'm curious now  )Cheers Jacko |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hello carl.h,
----------------------------------------Rosetta now keeps to a pretty good average re Boinc points .....A given machine will earn X an hour. I stopped reading most of the points arguments at Rosetta@home when they got so ferocious. How did it end up? I thought some people left the project because they felt so strongly against the proposed change?  Lawrence Added: Are we talking projects as in the Boinc sphere or the projects in WCG ? We are in this for the long haul. Consider both over a period of time with lots of different projects. And remember, I am hypothesizing that benchmarks for each type of core will vary from compiler to compiler in a way that is not necessarily closely related to the true optimizing ability of that compiler. [Edit 1 times, last edit by Former Member at Oct 30, 2006 9:06:49 AM] |
||
|
|
|