Just picked up a repair unit. Took 24 hours for the first owner to error out with 0.0 cpu time. So far it's running just fine for me, we'll know in about 5 hours if it runs successfully.

Error message for the first computer was -
Result Name: BETA_ c4cw_ target02_ 074906356_ 0--
<core_client_version>6.10.58</core_client_version>
<![CDATA[
<message>
process got signal 11
</message>
<stderr_txt>
Commandline = ../../projects/www.worldcommunitygrid.org/wcg_beta12_lmps_6.14_i686-pc-linux-gnu -screen none -in in.wcg.acc -var wcgsteps1 1000 -var wcgsteps2 10000 -var loop 0 -var restart 0 -var rinterval 100 -var ifile in.wcg.acc -var wcgseed 74906356

</stderr_txt>
]]>

My two betas are valid.

1) 3.34 Hrs vs ~5.89 avg on a C2D 2.0
2) 4.42 Hrs vs ~7.10 avg on a P4 2.8

Both with WinXP 32bit.

Nice improvements!

FWIW, Credit-wise, the difference is:
1) 14.37 GBCxH vs ~12.42 avg
2) 10.90 GBCxH vs ~10.26 avg

(GBCxH="Granted Boinc Credit x Hour")

Why would you think that? The few errors that I saw were resends and they all validated including my 10 original inconclusives. Lets rock'n'roll.

Sek,

Here is how credit work works for c4cw:

We have a table that stores the average credit per step completed in the workunit for each hr class (hr class for Linux, Windows, Mac).

Here are the current values for production:

28 256 0.00401590056974433 1166770
28 384 0.00674594873767008 6524044
28 770 0.00514894956281149 406545

Currently, all the workunits have the same number of steps per workunit.

256 -> Linux
384 -> Windows
770 -> Mac

This shows that Windows is being awarded .0067/.0040 = 167% of the credit that Linux is for completing the same work. This reflects the fact that Windows was running slower than Linux on the same workunits.

Points are awarded based on how much CPU time is used on a work unit. Since Linux, Mac and Windows are all different operating systems, a different compiler was used to generate the binary code for these application. In some cases the linux compiler produced much more efficient code than the Windows compiler and thus took less CPU time for a work unit.

On the 1:1 credit per work unit with no OS type in mind. That would not be fair to the different operating systems. Because of how the science application was compiled members should not be penalized for being told to run a curved route, when another OS was told to run a straight line to the finish line.

Also, the double points members were noticing originally had the previous beta values in it for credit per work unit. Thus for every result returned the running average changes by 0.1%. This is what I meant by "credit granted will correct itself back to normal". A new running average will be calculated once a majority of the work units running the new version returned their results. But in light of this, we are working on modifying the validator now which will actually create a new running average for the 6.14 science application separate from the 6.13 application.

The increase in speed you noticed on your machines is what we have seen on par with the entire beta windows improvement.

We are working towards making changes for the other science applications where there is a runtime difference between Linux and Windows.

Also, the beta work units were sent out with Zero Redundancy so there should only be a few more stragglers of work units returning.

-Uplinger

I need to work on switching my machines to Linux!