The intent of this post is not to start a flame war. Rosetta/Ralph/users are questioning methods for determining what claims are from standard hosts and which are from third party hosts. Note: optimized clients in and of themselves aren't bad, they just need to be properly compiled so that benchmarks are "inline" with the standard client when it comes to system types. To date I know of none who are "inline", although some come closer than others.

What follows is a spreadsheet of all the officially made Boinc CC's, which OS they were made for and release dates:




At the bottom is a list of known third party optimized apps. Weeding out the third party results isn't that tough, since many of the app names don't correspond with the "official list" as it pertains to OS. For Example;
Any result from 5.5.0 can be discarded as the official 5.5.0 was for Linux only and then only a developmental version (never released to the public as a "recommended" version. Also, I've yet to come across one in my studies).
Any result from 5.3.6 and Linux can be discarded since it doesn't officially exist. Neither does 5.2.12 with any OS, or 5.2.14 when used with Linux. Now it gets a bit more complicated, but applying the above filters will eliminate 75% of them. Trux and I believe some versions of Boincstudio have a calibration feature which allow the users to turn on/off the calibration, so a comparison of benchmarks to known standard boinc clients will help determine whether or not overclaiming is happening. The same is true with 5.2.11, 5.2.13, and 4.45, as they have both standard and optimized versions. My earlier study of the hosts of users who posted can be found here and by weeding out the easy ones shown above, what's left is a small sampling of many Hosts which use standard boinc benchmarks and by comparing the suspect one to the others, they can be weeded out.

To make it even easier, just weed out all the host info for all of them,including what might be valid benchmarks and you have a list of reliable benchmarks. Then apply the claimed credit formula and you have a pretty good example of what claimed credit per hour for each individual system should be. Now data needs to be collected on the number of models done per hour by those hosts and a reasonably accurate credit/model/cpu type can be established. Rosetta has access to the DB and is in the best position to do this as they "have all the data at their finger tips".

To determine Project Parity, simply take the remaining hosts, apply the claimed credit score and go with that. My Cross project comparision (an ongoing work) shows most projects grant nearly what the official claimed credit is supposed to be. Note: Seti and Einstien are still trying to figure out the best way to acheive parity and will be changing their credit systems.

[edit]overclaiming can also be done by editing XML files, so a comparison is always handy, but if you apply the above, you'll quickly begin to see what "normal" benchmarks look like and spot differences.

Not to worry a new plan for determining credit is already in testing at RALPH.

---

BOINC WIKI

BOINCing since 2002/12/8

That's what I'm talking about John, determining a method for calculating the best cobblestone/model possible and still maintain cross project parity.

I applied my own suggestion to my old data base, by eliminating results from any host that could possibly be overclaiming with software. Here's the result:







See how a range starts to appear for a given CPU? Now imagine if the project did this with the entire data base. I imagine they'd have a pretty good idea of what benchmarks and then claimed credit/hour for a given host should be. Then extract data about how many models/hour each does and make the adjustment.

Very interesting document, thanks for posting it. I see that see that overclocking by 43% only results in ~35% increase in credits/hour, guess the memory bandwidth is the limiting factor! I'm looking forwards to the new scheme turning up so I can finally compare my two boxes properly. Credit based on models processed is I think the correct way to go.

(Incidentally, I changed the CPU from a 3000 to a X2 4600 (overclocked to 2530MHz) on the 1st, so the most recent figures on the same host ID, 138077, will be a bit confused. As long as you collected the data before that date it should still be accurate).

Boinc version 5.3.19 was the first BBC/CCE version (I think this has been replaced by a 5.4.9-based version now, but don't know the timestamps for either, although you can download it from bbc.cpdn.org). I think both are windows-only but may be wrong.

---

out of curiosity, how come two of mine are different colours on there? is that to do with them running old BOINC versions? (they're both remotes so i haven't bothered upgrading them).
cheers
Danny

---

out of curiosity, how come two of mine are different colours on there? is that to do with them running old BOINC versions? (they're both remotes so i haven't bothered upgrading them).
cheers
Danny

Here's all your hosts in my DB:


I organized boinc versions by colors to make them easier to distinquish. You have a mix of Boinc Versions. The Sempron didn't have any completed results that I could check the Result ID on, so I used an "x".

does this help?

tony

this data was extracted roughly 18 days ago, so things may have changed.

Can i pick my own colors? How about BRIGHT lime ALIEN green! I really want to stick out on your spreadsheet!

---

SETI...Grown Men Looking For ET's That Dont Exist... Do You Dress Up Like Starwars Characters Too?

Odd, in that table, my "Evesham" node, (number 149), is still showing half Gig of RAM - I noticed this in an earlier table Tony posted. I have stripped that machine down, pulled and re-inserted the RAM sticks, and memtest86 assures me that there is 1GB in there, and that it is 100% good.

Perhaps like the processor type, (it is a P-IV Northwood core 2.533GHz), NT4 reports it weird?

---

Wave upon wave of demented avengers march cheerfully out of obscurity into the dream.

Odd, in that table, my "Evesham" node, (number 149), is still showing half Gig of RAM - I noticed this in an earlier table Tony posted. I have stripped that machine down, pulled and re-inserted the RAM sticks, and memtest86 assures me that there is 1GB in there, and that it is 100% good.

Perhaps like the processor type, (it is a P-IV Northwood core 2.533GHz), NT4 reports it weird?

Adrian, this is the same data, just mixed different. It's 19 days old now. It shows 1M now.

Memory 1023.11 MB

Ah, I assumed it was new data as it was dated yesterday. Cheers Tony.

---

Wave upon wave of demented avengers march cheerfully out of obscurity into the dream.