Obviously using an optimised client.

Trog Dog, whatever these 'special' BOINC compilations are called.

Here another one from the collection.....again 40 credits per hour claim....not so, unless 2 of them happen to meet in the same quorum......would WCG be able to tweak the distribution rules so that known 'Excessees' never meet up in the same 'team'? Maybe another rule, where the median can never be twice the lowest claim? In the example below that'd mean 42 as the award.


PS, i'll dig out the fixed credit discussion.....effectively, regardless doing it in 2 or 4 or 6 hours would get X for all, predetermined credit upon distribution. All benchmark tweaks would thus be nulllified with the BOINC open source agent.

Einstein@Home recently changed to this kind of system. Due to the nature of the science they perform (searching large data sets for patterns), they can easily predict in advance how long a given WU will take to complete compared to another.

Their science algorithm is very consistent, and the length of time required to process a WU is directly proportional to how much data (sky area) it searches and what sampling rate is used. They simply determine what a fair score for a certain WU would be, and then scale the scores for any other length of WU according to the relative workload. Right now they're only sending out 2 standard sizes of WU; a "short" WU is always worth exactly 21.48 credits, and a "long" WU is always worth 177.55, no matter how long it takes you to crunch them. Simple and completely fair; if your machine is twice as fast as average, you'll crunch twice as many WUs per day and get twice the total score.

But WCG doesn't have it nearly so easy. Their science apps are quite nondeterministic (for the current projects, anyway). The length of time required to crunch a WU isn't very predictable based on the initial data sent out; it depends very much on what's discovered while the WU is crunched. In my list of recently completed HDC WUs, I've got times ranging from 1:00:22 to 3:06:15. WCG doesn't know how much work a given WU will take to crunch in advance; they have to wait for us to tell them how long it actually took.

I've got an idea for a scoring system that would work well even for WCG, but it would require a lot of server-side coding to implement:

The project would designate one of their own in-house test PCs as a "calibration standard" system (preferably a very common, average system like a stock 3GHz P4). That system would crunch some WUs and log them into the DB as normal, with an additional "calibrated" flag set on the results. The score for these officially calibrated WU results would be based on some fixed, fair credits per hour rate typical of that PC on other BOINC projects.

The server would then take the results returned by the other 2 hosts assigned those WUs, and compare the time it took them to complete the WU to that used by the official standard host, to compute an official calibrated speed rating for them (which would be saved on the server along with the other per-host stats). These additional hosts would be granted the same credit reward as the official calibration system.

From then on, all results returned by those systems are also flagged as calibrated, allowing other, additional systems to have official calibrated speed ratings assigned in the same way. This would quickly pyramid through the system, assigning calibrated speed ratings to all systems. All hosts assigned a given WU would get the same credit reward, based simply on their official speed rating * CPU hours used * standard credits per hour rate for the original calibration PC.

If none of the hosts assigned a given WU are calibrated, the results would stay Pending Validation until one of them becomes calibrated (which then allows the others to be calibrated as well). To allow for changes in host speed, a given host's calibration could simply expire after so many days, causing it to be recalibrated again, much like BOINC currently re-runs the benchmark every 5 days.

I think something like this would work pretty well, and give fair scores to everyone without the current cheating and problems with inconsistent benchmark times on different platforms. But this kind of major recoding of the standard BOINC server software would take a lot of work to implement.

It's frankly not worth the grid tech's time to put so much work into this, so long as the current credit system is tolerable enough to not be chasing contributors away from the project. The current quorum system doesn't prevent cheating, but it dilutes the effectiveness of it down to where it's not really a big problem (yet, anyway; if lots of people started running "optimized" benchmarks, it could become one).

Cheating is a horrible problem with Rosetta@Home. With a quorum of 1 and no safeguards, people can basically claim any outrageous credit they want and get it, making their scores worthless. I wonder if that makes them less popular, or more; do more people leave in disgust than are attracted by the "opportunity"? It'll be very interesting to see how their volunteer count changes if & when they find a way to stop the cheating.

The current credit system is acceptable to me. ( I stopped using Truxoft, btw, when I discovered I was getting points I didn't deserve.) The cheating is diluted by WCG's quorum.

I think any system based on approximation of flops wiil draw complaints too because the approximation can be very innacurate. The only accurate flops method is to use the flops reported after the WU is crunched. If the flops could be hidden or encrypted in the crunch report returned to WCG so that cheaters could not tamper with and inflate the flops then we might have an honest and accurate credit system. But I think cheaters would eventually crack the system and the cheating would start again.

Weighing the benefits against the efforts required to end cheating, I vote for leaving the system at WCG the way it is.

Most of us admit the credit system is important and arguable. Some of us think the system here at WCG is good enough and that a system like R@H's new system will not bring lasting benefits. Some of us feel there are much more important issues that need to be addressed first.

From R@H we will learn that the new system stopped the cheating for a while but eventually the system was cracked and the cheating started all over again. Show me a credit system that cannot be cheated and I will support it 100%.