That is not an improvement. It's just a new way of committing the same mistake, the mistake of allowing host computers to evaluate their own performance. That system will always be cheated.

Furthermore, if a project can devise a test WU that will way of predict the performance of all hosts when those hosts run the real WUs, then surely the project has enough knowledge of its science app and WUs to just assign fixed credits on the host, as many projects are doing today.

Any system that allows hosts to evaluate their own performance will be cheated and anybody who believes otherwise is delusional. You can add encryption, hashes, whatever you want and they will be broken too. Modern encryption and hashing is secure when there are zero to very few clues into the encryption/hash. Unfortunately, the exchange of WUs and results will provide cheaters with more than enough clues to break whatever security is built into this supposed "improvement". It might take a lot of computing power to break it but such power is easily obtained today... just setup a BOINC project, advertise it, hand out ridiculously high credits and wait... soon you will have thousands of fools merrily crunching whatever you ask them to crunch, no questions asked, just wash away any curiosity with rich lather of credits, propaganda and censored forums. Sheeples will do anything when you know how to use them and how to herd them.



Do convicts design and build the prisons that hold them? Ridiculous!! Open source security, yeah right!! Has Merck/Frosst created a naivite pill?
.

Read the OP's post. That's not the problem.



Illogical and just plain fear mongering and FUD slinging.



If you're happy with the oars you got then row. Those who get broken oars will fix them, if they choose, using advice that travels fast around the 'Net, like a FUD purging.
.

Aplogies for resurrecting this thread after so long, but I just noticed something.

Just now I looked at my pending credit and noticed that the claimed credits all seemed a higher than I am used to seeing (almost x2). I looked at the benchmark scores currently stored in Client_State.xml and these seemed to be almost double what I remember. I kept a copy of Client_State.xml and re ran the benchmarks. This produced figures more in line with what I usually see.

Pasted below is the erroneous High benchmark:
<p_fpops>4675852066.715012</p_fpops>
<p_iops>8225251508.120034</p_iops>

This is the more usual score:
<p_fpops>2041665416.451956</p_fpops>
<p_iops>3181556128.139779</p_iops>

I originally assumed that the variation in Benchmark was only one way - ie sometimes lower than the "true" figure due to processor being interrupted during the benchmark run by critical windows system components. Now either this was a one off error, or the variation can be a double edged sword - penalising or over rewarding. I have re run the benchmarks 5 times now though, and I get the "expected" (lower) results consistently. It seems the fates are only generous once in a blue moon!

I agree with knreed that adopting ones own benchmark scores would be a questionable practice - even if done using fair figures. Would it be fair for me to use my occasional high score? No. In any case, I should see my batch of high credit claims ironed out by my quorum companions. Usually my claimed and granted scores match well.

I only started looking at benchmarks because I had forked out £80 for a new processor - from a P4D 830 3GHz to a P4D 950 3.4GHz. The clock speed increase is small, but I was assuming that increased cache of 2x2MB from 2x1MB would also make a big difference. It did not. The increase I got was only about 3.4/3 = 13%

13 % - Pah. So, how to make the lazy little Intel g*t work harder then? Overclock! I increased FSB from 800 to 957 which puts core speed up to 4.07Ghz. So increase now is 4.07/3 = 36%. Thats better!

But my CPU temp went up 10degC !! Reduced Processor Voltage by 1 notch - 0.0125V and it went back down. Now at 55-60degC.

System has now run like this 24/7 for about 6 weeks without a single crash and without a single errored workunit.

I think I will stick there. I have heard tales of Overclockers really turning the wick up on Intels and plumbing their processors into their central heating etc - but I am sure the little chap will only sulk if I push him any further.

I have never used an "optimized copy" (whatever that is), nor have I ever manually intervened in the Benchmark scores. I get official BOINC updates directly from the website. I dont run Linux.

I only ever saw 4675/8225 once, having first noticed the high credit claims. A Benchmark of 2032/3179 is typical now (after proc upgrade).
SiSoft Sandra gives 18148 MIPS and 14983 MFLOPS, which bear no resemblance to the BOINC figures, although this does not matter.

My last BOINC upgrade was 5.8.11 to 5.8.15. This blip in benchmark score happened at some point after this. I was just curious if it was a rarely seen bug (relating to dual core processors) in BOINC and if any one else had noticed anything. I am still trying to get it to do it again.

I have just installed the new BOINC version 5.8.16 (from 5.8.15), in the hope that software updates improve things

Back a step! Whereas I was getting benchmarks of 2032/3179 fairly consistently with v5.8.16, after the upgrade I now get mostly figures of 2047/1773, and OCCASIONALY (1 run in 5) I get 2048/3225 - similar to what I used to get.

At least the Floating Point figure is consistent, and maybe it is only this that matters??