It's not for me. I already know.

In a nutshell: Because linux machines have lower benchmarks for the *same hardware*, it is required to use a boinc manager that increases the benchmark to make it even to the windows machines.

What you did is an unsupported option, not a requirement.

This is a prime example of 'skewing' points so that problems are difficult to identify and hence difficult to fix. If everyone used the client in its standard form this type of thing could be seen easily and fixed. Your methods muddy the waters, this could have been done years ago.

BTW. This problem was brought to the attention of WCG, tested as valid by them and WCG contacted BOINC to correct it in BOINC 5.8. That information is contained in the first post of this forum.

The more obvious answer was given to Dagorath when he made the same claim but its content was removed as being offensive so I wont repeat it here.

Cheers. ozylynx

Hello Zombie67

I understand your points here. In fact there was 6 days notice given. Not 7 but not 0 either. Please check the times and dates of the two forum announcements.

The forum is the official vehicle for such things and it is our responsibility to check these things out. This particular thread, containing 'copies' of the official news notice, was kept on the front page of the 'recent threads' menu for the vast majority of those 6 days.

{Edit} Just saw your latest post. Please type "KISS method" into the search box and read my thread on benchmarking. Note the dates too.


To save another post:
Further to comments on slow machine benchmarks... I might be on to it!!
The formula used for benchmark calculation gives equal weight to Gflops and Giops. The science , I think, is more dependent on floating point than integer operations. On my machines the Gflops fairly closely represent the clock speed of the CPU as a proportion. The Integer ops do not. So a legitimate benchmark is obtained giving too much weight to the Integer ops which are not as critical to the completion of the WU. Thus the older slower machine takes longer than its indicated benchmark to complete a WU and claims more points. I have found that a benchmark calculation using Gflops scores only will in fact produce a proportion which is much more accurate.

Hope this helps.

Cheers. ozylynx

I always thought the way point are awarded in most BOINC projects are flawed. The problem is thinking that one benchmark can reflect the computers capabilities for any project. Ozylunx gives one example of how the science application can differ from the benchmark routine.

If one want to have a really fair points system, the benchmark should be individual for each science application, and sent out by the science project. It could be the time to crunch kind of a "mini work-unit", thus reflecting the true speed of the science app. An alternative for some projects could be to ignore benchmarks altogether and do it the CPDN way. That could be a problem for projects where the computing time can vary a lot between work-units though.

Optimised science applications is a problem in the context of awarding points (although they are good for the science - more work will be done). If benchmarks are unchanged, people with optimised science will claim lower points. That may affect people using a standard science app, leaving them with too few points. OTOH, people with optimised benchmarks (BOINC clients) will claim too many points if they are using a standard science app. Now if I use an optimised science app in one project and not in another, there really is no solution with the current design of one generic benchmark.

I myself was for a while running WCG with an optimised BOINC client. That was because I started out with SETI and optimised science apps. I then started cruching CPDN, which was no problem since they ignore the benchmark results and just award points per trickle. Then however I also added WCG, and it became a problem. I kept the optimised BOINC client though, because I crunched less WCG than SETI, and I reckoned that I did the least damage that way.

Now I have left SETI (I think there are more important projects to work on) and am therefore running a standard BOINC client.