AMDave
02-17-2013, 08:49 PM
New LLR credit function
Your voices have been heard! Well, we always heard you, but this wasn't the easiest thing to fix. Anybody who has ever run any LLR project here probably noticed how seemingly random the credit was. Without going into detail, the variation was due to using the BOINC benchmark * CPU time method for calculating credit, which has lots of problems. I've written a new credit function that is based soley on the complexity of the problem, so credit will be a lot more consistent than before. We'll be turning the new credit system on one project at a time. You may notice that some projects will get more credit than before, and some may get less. The adjustment is to make it so that you get the same credit/hour regardless of which project you're crunching. Currently, the new system is being used only for SGS. If nothing really bad happens, we'll start turning it on for the other projects. SGS used to have a fixed credit. It now varies slightly according to the size of the number. VERY slightly -- the difference is about 8 places past the decimal point, so you won't notice the difference, but it's there. This should correct the following inequities with the old credit function:
Identical Windows and Linux boxes scored differently on the BOINC benchmarks, and therefore got different credit for the same amount of work.
Sandy and Ivy Bridge CPUs got LESS credit for being faster. Your wingmen would affect your credit, so if THEY have a Sandy Bridge CPU, you get penalized because their computer is fast.
I'll keep you informed as the other projects are moved to the new system.
More... (http://www.primegrid.com/forum_thread.php?id=4862)
Your voices have been heard! Well, we always heard you, but this wasn't the easiest thing to fix. Anybody who has ever run any LLR project here probably noticed how seemingly random the credit was. Without going into detail, the variation was due to using the BOINC benchmark * CPU time method for calculating credit, which has lots of problems. I've written a new credit function that is based soley on the complexity of the problem, so credit will be a lot more consistent than before. We'll be turning the new credit system on one project at a time. You may notice that some projects will get more credit than before, and some may get less. The adjustment is to make it so that you get the same credit/hour regardless of which project you're crunching. Currently, the new system is being used only for SGS. If nothing really bad happens, we'll start turning it on for the other projects. SGS used to have a fixed credit. It now varies slightly according to the size of the number. VERY slightly -- the difference is about 8 places past the decimal point, so you won't notice the difference, but it's there. This should correct the following inequities with the old credit function:
Identical Windows and Linux boxes scored differently on the BOINC benchmarks, and therefore got different credit for the same amount of work.
Sandy and Ivy Bridge CPUs got LESS credit for being faster. Your wingmen would affect your credit, so if THEY have a Sandy Bridge CPU, you get penalized because their computer is fast.
I'll keep you informed as the other projects are moved to the new system.
More... (http://www.primegrid.com/forum_thread.php?id=4862)