Eric J Korpela wrote: > On Wed, Sep 30, 2009 at 9:42 PM, Paul D. Buck <[email protected]> wrote: >> Yep, that is part of the definition... how is Eric's deflationary >> multiplier that debases this an improvement? > > You continue to claim that the credit multiplier is deflationary. I > don't see how you reach this conclusion. The multiplier is remarkably > stable, and any instability almost entirely due to variations in how > the data is analyzed rather than variations in the host machines or in > the BOINC clients that they run. It's certainly a far more stable > means of granting credit than the "benchmark x time" method since its > the median of all machines rather than the lower claim of two machines > or the median claim of three.
The median of a large number of hosts will indeed 'tend' to be stable. That's why medians are used (and abused). However, will that 'stability' survive a long run of same-angle-range s...@h WUs followed by a jump to another long run of same-angle-range s...@h WUs that are a different angle? Eg: A run of VHARs followed by VLARs or vice-versa? Or even if the s...@h FLOPS counting is updated (changed)? What happens then to the median and the credits adjust? And there is the vulnerability of when GPUs eventually push into the median... The present script does indeed /appear/ to be working well for the present conditions. But, "what happens next"? And can we say anything 'meaningful' about the present credits? I'm sure that many use the credits as a vague performance indicator. It sure would be good to be able to use them as an absolute (and historical) performance measure... Regards, Martin -- -------------------- Martin Lomas m_boincdev ml1 co uk.ddSPAM.dd -------------------- _______________________________________________ boinc_dev mailing list [email protected] http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address.
