Nicolás Alvarez wrote: > I have now posted it on: > http://www.boinc-wiki.info/User:Nicolas/Credit_scenarios > [...] >> >> Has anyone checked how this new design helps with the two (three?) >> credit scenarios I posted on the thread "[boinc_projects] Catastrophe >> is looming; desperately need advice" on Jul 14 2009? >> >> I think I should put them somewhere more permanent... >> >> -- >> Nicolas
Very good clear scenarios there and a good example of one real world project being destroyed as a result. "Fair" and _consistent_ credit is _important_ Your examples are all an example of there being no consistent real-world (hardware) reference on which to base the credits. Hence, there is no reliable 'anchor' for the credits value. Nor is there any indication or any feedback to indicate how efficient or inefficient an application might be. A project is forced to set 'arbitrary' numbers unless there is some real-world reference. Having said that, the present proposal appears to be a good thoughtful step in the right direction. Myself, I still feel that the "Cobblestone" definition[1] is flawed due to the disparity between the hardware cost of floating point operations versus integer operations in real world hardware whilst Cobblestones reward both equally and so to the great disadvantage of the true real-world costs floating point. My view is that we need a new definition whereby: Cobblestones_v02 = kc ( kw * Whetstones + Dhrystones ) where: kw reflects the relative performance for the Dhrystones[2] benchmark compared to the Whetstones[3] benchmark such that for the most common (modal) hardware, kw * Whetstones = Dhrystones. Special note: Note the use of " *modal device* " rather than "median performance". kc is used to give an initial approximate parity to the existing Cobblestones credit for an arbitrarily chosen fixed reference date. The Cobblestones_v02 MUST be calibrated on a "golden" reference piece of hardware. Also known as a "Etalon" computer from a previous (long) discussion. A second problem still is how do we credit projects that DO NOT have numerical calculations as their primary (most significant) operation? For example, projects that merely require disk storage, or network availability, or low resource real-time monitoring? (Cobblestones_v03? Or a 'supplementary' Bonus_Cobblestone?) Perhaps we really do need a second parallel credits scheme that is deliberately differentiated from the credits-whores power-crunching Terra-Cobblers machines? Without a *fixed reference* we're just shooting at clouds and recent scenarios will continue to be repeated... All not to be confused with: http://en.wikipedia.org/wiki/Cobblestone nor http://en.wikipedia.org/wiki/Shoemaking (and the associated slang...) Regards, Martin [1] http://en.wikipedia.org/wiki/BOINC_Credit_System#Cobblestones [2] http://en.wikipedia.org/wiki/Dhrystone [3] http://en.wikipedia.org/wiki/Whetstone_(benchmark) [4] http://www.boinc-wiki.info/User:Nicolas/Credit_scenarios -- -------------------- 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.
