Hi Otis,
I've been with WCG since 12/25/08. My (anal) accounting backround has hassled with this same situation. By Trial/Error and (probably way too much..) attention to completion times/points/etc. I've managed (i think) to maximize the output. Single processor machines up to 1.6Ghz seem to thrive on Rice and Muscular Dystrophy (with programmed cut off points), while it takes a dual processor w/1+Gig RAM to really 'get down' on the CEP, and other more complex studies. I have a 733Mhz and two old 566Mhz eMachines(both donations) that I increased RAM to the max (384meg) that crank out Rice units 21 of 24hrs regularly. Should hit Ruby-Rice by end of October. My 3 dual processor machines, meanwhile, work out on MuscDystrophy, Cancer, ChildCancer, CEP.
Just thought you'd like to know. I concur with your idea of having BOINC-WCG internally compute the best study for a certain machine. Would SAVE a ton of math/observation on our part.
Regards and Happy Crunching,
JR (ol43)

Wow, I had given up on actually getting a response to this, then i run across it today and have two!

Ok, so I believe this is the log WCG makes on each client computer if I am not mistaken. (feel free to correct me at any time) But what would need to be done to implement what I am suggesting would simply to put in some extra fields and make it look like


<host>
<total_credit>16393.197724</total_credit>
<expavg_credit>0.076480</expavg_credit>
<expavg_credit_HCMD2>0.076480</expavg_credit_HCMD2>
<expavg_credit_IADS>0.085645</expavg_credit_IADS>
<expavg_credit_HFCC>0.086415</expavg_credit_HFCC>
<expavg_credit_HPF2>0.065891</expavg_credit_HPF2>
<expavg_credit_FAH>0.074365</expavg_credit_FAH>
<expavg_credit_CEP>0.015243</expavg_credit_CEP>
<expavg_credit_DDDT>0.046538</expavg_credit_DDDT>
<expavg_time>1158057017.453430</expavg_time>
<p_vendor>AuthenticAMD</p_vendor>
<p_model>AMD Athlon(tm) XP 1700+</p_model>
<os_name>Linux</os_name>
<os_version>2.6.9-34.EL</os_version>
<create_time>1130851159</create_time>
<rpc_time>1152655444</rpc_time>
<timezone>-21600</timezone>
<ncpus>1</ncpus>
<p_fpops>759071089.284725</p_fpops>
<p_iops>1297318240.976990</p_iops>
<p_membw>1000000000.000000</p_membw>
<m_nbytes>528044032.000000</m_nbytes>
<m_cache>262144.000000</m_cache>
<m_swap>1073733632.000000</m_swap>
<d_total>39237181440.000000</d_total>
<d_free>31422160896.000000</d_free>
<n_bwup>38598.000000</n_bwup>
<n_bwdown>0.000000</n_bwdown>
<avg_turnaround>180410.906113</avg_turnaround>
<credit_per_cpu_sec>0.001190</credit_per_cpu_sec>


I believe it is WCG program that accesses this information correct? (or is it Bionics?)
If it is the prior, the code simply need to add some conditions to run the most efficient one. Like this computer for example would request HFCC first, and if no work is available or it is on a lower priority, request IADS. It would just keep going down the line until it got some work. In this example, the computer above seems to have a large problem with CEP WUs as it only has a 0.015243 points per hour output, so would request that last.

This could be worked in with the existing structure of WCG's servers, and only ask for the most effective pairing in the queue. This situation would still clear the queue, and would be more effective at scientific thorough put than the current method. (all assuming points are an indication of the amount of the science done.)

If all of the statements above are correct, I do not see that this would be that difficult. If you would point me in the right direction (and I have access to the code - is it public?) I would gladly try to implement something like this myself and then send it in for WCG approval - but I have a feeling WCG would not do that...