The Rosetta app has always run fine before but the method for calculating credits is very flawed and I will not believe that these processors are in anyway one third the speed as the athlon.

I definitely understand your feelings, but if I compare my numbers with those of a Power Mac 11,2 then my machine crunches around 2,5 times that of the mac per core.

WU time decoys time/decoy
Power Mac 11,2: t322 13340 s 3 4450 s
X23800@2.2 GHz: t322 41500 s 12 1800 s

I can do the same for quite a number of other WUs for the same machine and it turns out the same. It is as Ananas said, it is a question of optimizing the science-app code for a specific processor architecture. So the system itself is not flawed, but it turns out that "counting real work done" is bad for some machines which I sure do not think is done intentionally but just a matter of cold numbers.

The Rosetta app has always run fine before but the method for calculating credits is very flawed and I will not believe that these processors are in anyway one third the speed as the athlon.

I definitely understand your feelings, but if I compare my numbers with those of a Power Mac 11,2 then my machine crunches around 2,5 times that of the mac per core.

WU time decoys time/decoy
Power Mac 11,2: t322 13340 s 3 4450 s
X23800@2.2 GHz: t322 41500 s 12 1800 s

I can do the same for quite a number of other WUs for the same machine and it turns out the same. It is as Ananas said, it is a question of optimizing the science-app code for a specific processor architecture. So the system itself is not flawed, but it turns out that "counting real work done" is bad for some machines which I sure do not think is done intentionally but just a matter of cold numbers.

If its counting the "real work" then its not "bad" or a flaw, then it shows that the rosetta app needs to be "fixed" to work at the correct speeds on a mac, depending on how widespread this is.

---

How are WUs with 12 hour run times comparing between old and new credit system using the standard client on Macs? And if it's just some of the standard client Macs that are getting less than they would have under the old credit system - then perhaps visiting one of the Mac DC groups for tips for streamlining the system to run Rosetta would be in order. The folks at MacNN (or other collections of Mac experts) might be able to identify programs running in the background that would lessen model/decoy production rates.

At Distributed Folding, one of the arguments against Altivec optimizations being hand coded was the fact that the code spent 90% of the time in a loop that wasn't Altivec ready. If Rosetta has a much larger portion of the code that could be optimized for Altivec, and if the compilers used support Altivec optimization, then now would be a good chance to request a test run of the Rosetta client compiled with the Altivec switches turned on.

On the other hand, if you're part of a team that is running multiple projects - you might see if someone on a more Altivec optimized project has systems that would work better here - and trade. Fold for them on the Altivec optimized project, and have them fold for you here on Rosetta.

---

On the other hand, if you're part of a team that is running multiple projects - you might see if someone on a more Altivec optimized project has systems that would work better here - and trade. Fold for them on the Altivec optimized project, and have them fold for you here on Rosetta.

This relates to the whole idea of finding out what runs best on each computer - different algorithms in rosetta, or (sorry!) even different DC programs on boinc. It really would be a shame if the processors/architechture in some pc's let it do certain projects faster than others (compared to a baseline of some sort) and we didn't capitalize, atleast in rosetta, by feeding it atleast 50% of the time the ones that are "optimized" for that type of computer! (other 50%, or less, can be testing / running other things that may be urgent)

---

When will the new credit system become official, and exported as such to the BOINC stat sites? Is there a target date?

Thanks.

---

Reno, NV
Team: SETI.USA

When will the new credit system become official, and exported as such to the BOINC stat sites? Is there a target date?

Nothing in stone yet. Still working on it and don't want to promise a given day or timeframe.

---

Thanks to all that are truly concerned. And to Ananas you don't know me and I don't know you so there is no way you could offend me I would also never be offended by or try to offend anyone from another country or that speaks another language. In fact if you can even a little speak two languages you are better than me in my book. (unless you wrote the code that deliberately limits my machines) Yes I sad if someone writes code the deliberately restricts the use of all the power of a large part of this DC community you are insulting all those users. If that is Mac user or Linux and yes even Windows users with their SSE and stuff. I may not be a programmer but it would not be much more than a week to add these optional features especially if you enlist the help of the people that recompiled all the optimized clients for SETI. They would probably help just for credit instead of what happened to them at SETI. I say let them optimize then verify in Ralph and let everyone run optimized clients as standard.
It is however too late for me as I have already hit that "Won't get new work" button on all my machines.

Thanks again,

SETI ended at 267,477.16
SETI Classic WU 42,282
Einstein ended at 8,458.52
Rosetta will end around 380,000

Total around 655,000

---

... The folks at MacNN (or other collections of Mac experts) might be able to identify programs running in the background that would lessen model/decoy production rates. ....

I think they might be able to do more. Even if Rosetta is not freesource, volunteer support from people who know how to optimize a client might be welcome.

Einstein had quite a lot of success with their well-known volunteer assembly language wizard, SIMAP has done similar, why should not Rosetta do something like that too and ask the Mac professionals to compile vectorized clients for optional download?


edit : Same idea but you found the words faster than me ;-)

I may not be a programmer but it would not be much more than a week to add these optional features especially if you enlist the help of the people that recompiled all the optimized clients for SETI. They would probably help just for credit instead of what happened to them at SETI. I say let them optimize then verify in Ralph and let everyone run optimized clients as standard.

Unfortunately it's not that easy (if it were - they'd do it - they're desperate for more processing power!). It's been shown that SSE etc probably wouldn't be very useful for the Rosetta algorithms. I'm sure there are optimisations that can be made, and I'm sure they will be.

HTH
Danny

---

If anyone is interested, you can gain access to the Rosetta source code by working with the U of Washington licensing folks. Suggestions on how to improve the code would be forwarded to the lab, and if accepted, worked into a future release. Here's the info to get started:

http://depts.washington.edu/ventures/UW_Technology/Express_Licenses/Rosetta/

Digital Ventures
UW TechTransfer
4311 11th Avenue NE, Suite 500
Seattle, WA 98105-4608
Phone: (206) 616-3451
Fax: (206) 616-3322
Email: license@u.washington.edu

---

If anyone is interested, you can gain access to the Rosetta source code by working with the U of Washington licensing folks. Suggestions on how to improve the code would be forwarded to the lab, and if accepted, worked into a future release.

This must be a great help for this project

If anyone is interested, you can gain access to the Rosetta source code by working with the U of Washington licensing folks. Suggestions on how to improve the code would be forwarded to the lab, and if accepted, worked into a future release.

This must be a great help for this project

I hope someone picks this up and runs with it. I will be watching from the sidelines to see. If someone does have an optimized client to try that is legit I will try it.
I will need one for G3, G4 and G5.

---

I have gone through that process with the licensing folks and have now access to the source. Mats Petersson and Leonard McGuire have looked over the code and it's at least not a trivial task to optimize for SSE. However they found many places where optimizations might work (not only SSE) but it would require some work. If anybody is interested in that kind of inspection or would like to look for optimization for PowerPCs drop me a note:

joachim@iwanuschka.de (I'm aware of mailbots and have my countermeasures).

edited because of incomplete post.

I have gone through that process with the licensing folks and have now access to the source. Mats Petersson, a

This is great and I may not have to shut down after all.

---

Question : Will this type of optimisation cause problems again ?

Not trying to cause trouble, just looking forward.

---

Not all Czech`s bounce but I`d like to try with Barbar ;-)

Make no mistake This IS the TEDDIES TEAM.

Question : Will this type of optimisation cause problems again ?

Not trying to cause trouble, just looking forward.

At the moment we are just looking for opportunities. If something promising is found it is forwarded to the baker people who decide. Usually optimisations don't cause any problems but every code change is a risk of introducing a bug.

I think it would be accurate to say tralala is working to find suggestions to submit to the project team for improving efficiency. I doubt he plans to release his own optimized client, in fact I'd bet the license agreement precludes that. But such efforts could make the Bakerlab distribution crunch more models with less time.

This approach will help avoid confusion within the user base, and assure there's only one client, the one distributed by the project. This is best, as they are always improving the approaches taken to study the proteins, so if a separate optimized client were produced it would quickly be obsoleted when Bakerlab releases their next new science changes.

---

Add this signature to your EMail:
Running Microsoft's "System Idle Process" will never help cure cancer, AIDS nor Alzheimer's. But running Rosetta@home just might!
https://boinc.bakerlab.org/rosetta/

Thanks for clearing that up Feet1st ;-)

---

Not all Czech`s bounce but I`d like to try with Barbar ;-)

Make no mistake This IS the TEDDIES TEAM.

Dr. Baker posted a new journal entry 8 hours ago, and here is part of it with bold by me for emphasis:

On the credits front: we have decided to use the average amount of time for producing a structure over all rosetta@home runs for a particular work unit to determine the amount of credit to be awarded for each structure produced for that work unit. so, for example, let us suppose that all rosetta@home computers on average took 1 hour to make 1 structure for a given work unit, and that this corresponds on average to 10 credits using the standard boinc accounting scheme. Then each computer gets 10 credits for each structure returned--a fast computer might be able to do 3 structures in an hour, and get 30 credits per hour, wheras my old slow laptop may require 2 hours to make a single structure so I would only get 5 credits per hour. I think everybody will be happy with this approach in the end, even though nobody may be very happy with it initially (I must emphasize that, contrary to some statements on the boards, no individuals or groups had any more influence on our strategy than any other, so I hope this issue can be laid to rest). I am sorry that the switch to the new system has generated so much conflict, I really didn't anticipate this, and I'm sorry that my attempts to calm things down only made things worse (again, I will post from now on only in this thread and only in this forum). In any event, we are quite set on the new credits plan, which we think will be better for everybody, and please hold off on comments or suggestions for two weeks or so until we all have a clear picture of how things are working--we do not need or want new suggestions at this point. David Kim will be posting the definitive description of the new system tomorrow on the boards.

Hopefully, we can post questions about the new system in David Kim's topic. I have several already. :-)

I think it would be accurate to say tralala is working to find suggestions to submit to the project team for improving efficiency. I doubt he plans to release his own optimized client, in fact I'd bet the license agreement precludes that. But such efforts could make the Bakerlab distribution crunch more models with less time.

This approach will help avoid confusion within the user base, and assure there's only one client, the one distributed by the project. This is best, as they are always improving the approaches taken to study the proteins, so if a separate optimized client were produced it would quickly be obsoleted when Bakerlab releases their next new science changes.

Should'nt that be application if we want some real optimization here? Dunno about you, but that really confuses me ? Dunno how all of that can be done with the client ?

---