[Shaktai]

Rosetta@Home is another Life Sciences project similar to Predictor, but taking a different approach to Protein Prediction. The software has been around for a few years and has had some impressive success. The goal of the project is to further help refine the software. -- It is non profit from the University of Washington. The software is free for academic science research institutions and is also used by some for profit pharmaceutical companies but they have to pay for it (proceeds go back into research and development) It has had some impressive wins in the CASP competitions.

The project is still in "beta" status, but is already delivering results. A lot of information is available on their website. There is of course a Mac Client. If you are one of people who doesn't feel inclined to help crunch on Einstein or SETI, this is a worthwhile project to look at.

The Mac app isn't optimized, but is still reasonable. It runs well on older G3s also. The work units do not require multiple validations from other computers. Therefore, if your work unit has no errors, you will get your credits right after it is submitted.

KNOWN ISSUES:
1.) Checkpoints are far apart. If you switch between projects, it will fall back to the last checkpoint on restart. Sometimes it seems to forget the time completed, especially if you are switching with Einstein.

2.) The first 75% goes quickly. However the process is two step. The first 75% is coarse calculations, and after that the calculations become much more complex (refined) and takes a lot more time, so the percentage is a reflection of work unit completion, not time to completion.

3.) If you are running multiple projects, you will probably need to increase the time between project switching from the default 60 minutes to 120-180 minutes. The slower your machine, the more time needed, otherwise after 75% you will never complete to the next checkpoint which is 83.33%.

The problems will probably all be addressed in future app versions. The development team seems to be very responsive. Windows users already have an optimized app that is very fast.

Team MacNN is present and accounted for by yours truly, and currently ranks #6 out of 118 teams (and still climbing). As always, I am a glutton for "beta testing" punishment.

It is beta though, has no graphics, and has some quirks.

If you want a safe and stable project that makes the best use of G4 and G5 processors, I would recommend Einstein@Home right now. Their altivec enabled app really screams on G4's and G5's. Predictor will need some crunchers again when it starts back up in a couple of days, so that we can hold our #1 spot there. (Predictor is another good project for G3 macs also).

If you are adventuresome and don't mind beta testing, give Rosetta a spin if you aren't already heavily committed to another project. Just remember to increase the time between project switching under General Preferences if you are running multiple projects. And remember it is BETA. Expect some problems along the way.

[Shaktai]

David Baker, the lead scientist for Rosetta@Home posted the following in their forum.

All of your help is doing wonders for our research! We will be posting the results of the current round of computations in the top predictions section when David Kim returns from vacation in London early next week.

However, we obviously haven't been able to solve all of the problems that have come up. David and I were thinking that we might be able to enlist the coders among you to make things even better. We could make the rosetta c++ code available through this site, and you could help us resolve platform specific issues and maybe make the code even faster. For anybody who is seriously interested and becomes more involved, we have an annual rosetta developers meeting every summer in the washington cascades which is a lot of fun.

Would anybody be interested in seeing the code, and helping to track down problems? There would be a very simple click on license agreement which exists because the annual meeting and some of the code development is funded by licensing the software to industry, so we have to require that the code not be used for commercial purposes.

Also, are there suggestions of things we should add to the home page?

Once the team meets their goals for Einstein, this project looks very interesting. An optimized Mac application would do wonders.

[Shaktai]

A small handlful of individuals have joined me on Rosetta. Currently the Team is #7 in Recent Average Credit and still climbing and #7 overall out of 169 active teams. We'll try to hold the line, while the majority of the Team MacNN advances in Einstein. Barring any "big" teams coming on board, I think we can make it to #5.

Pretty stable for a "beta" project. Occasionally has problems switching with other projects though. The project team is also very similar to the excellent Einstein project folks, providing great feedback and information.

[reader50]

Looks like we will have at least 10 credited members before too long. It would feel odd to arrange stats for a team-of-one, even with Shaktai's production.

If we hit 10 by this weekend, I'll start looking into stats.

[Shaktai]

Originally Posted by reader50

Looks like we will have at least 10 credited members before too long. It would feel odd to arrange stats for a team-of-one, even with Shaktai's production.

If we hit 10 by this weekend, I'll start looking into stats.

Well, we show 10 on the team, but only 7 have returned work for credit. I hope no one is giving up too easy. Rosetta is still having problems playing nice with other projects though, but that should be fixed soon.

Two significant problems right now, but they are being worked on.
1.) Getting stuck on 1%. The fix is to quit and restart.
2.) If running more then one project, getting stuck at 83.33%. This is the point where the calculations get more complex, and it takes more time to process. The fix is to increase the time between project switching. For all but the fastest computers that means 90-180 minutes. This is so the computer can reach the next stage before switching, otherwise it resets back to 83.33 on restart.

Good project othewise with a proactive development and science team. The project will be down to upgrade the server to BOINC 5.x on Wednesday (24 hours is planned). Now is a good time to increase the cache to 2 days in order to keep working.

[Shaktai]

Rosetta @ Home is officially out of BETA. The project is open to the public and appears to be stable. They have also upgraded to the latest version of BOINC 5.x. (and it was a very smooth transition).

[buggsuperstar]

I'm very interested in this project. But I'll give a little more time to Einstein first before jumping over to Rosetta.

Count me in.

[Shaktai]

Originally Posted by buggsuperstar

I'm very interested in this project. But I'll give a little more time to Einstein first before jumping over to Rosetta.

Sounds like a good idea. There is still progress to be made on Einstein.

Rosetta still does not have a 10.3.9 compatible app, and Tiger is still required for Macs. However David Kim posted today that he is working on one, and just needs to test and benchmark it. Hopefully 10.3.9 users will soon be able to join the effort. Until then, Einstein with its altivec enabled app is a good place to put your efforts.

[Snake_doctor]

Originally Posted by Shaktai

... 2.) If running more then one project, getting stuck at 83.33%. This is the point where the calculations get more complex, and it takes more time to process. The fix is to increase the time between project switching. For all but the fastest computers that means 90-180 minutes. This is so the computer can reach the next stage before switching, otherwise it resets back to 83.33 on restart. ...

An even eisier fix and one that does not take time from other projects is to keep the apps in memory. Unless you are short on memory this should have no effect on your system or it's performance on other tasks.

Regards
Phil

[99.94]

Originally Posted by Shaktai

Rosetta @ Home is officially out of BETA. The project is open to the public and appears to be stable. They have also upgraded to the latest version of BOINC 5.x. (and it was a very smooth transition).

I have been folding for a long time at Folding @ Home (different name, different team). Things over there have been pretty unsatisfactory for Mac G4 users and eventually I decided to move on.

I am with Einstein @ Home now. The Einstein project has the potential to make a significant contribution to our store of knowledge and I am quite happy to stay for the short to medium term.

It might be a legacy of a long exposure to F@H, but my preference would be eventually to go to a life sciences project. Rosetta @ Home fits the bill, but it would be with one big proviso: efficiency is important & the Mac client would need to be optimised to exploit Altivec and to be broadly comparable in output to similarly spec-ed machines using other platforms. E@H is a good model in that respect.

[Shaktai]

Originally Posted by 99.94

It might be a legacy of a long exposure to F@H, but my preference would be eventually to go to a life sciences project. Rosetta @ Home fits the bill, but it would be with one big proviso: efficiency is important & the Mac client would need to be optimised to exploit Altivec and to be broadly comparable in output to similarly spec-ed machines using other platforms. E@H is a good model in that respect.

Einstein is a good project, and it does provide an altivec app direct from the project developers. Plus it is a big push for team to advance in right now.

Rosetta is a much newer project, but they are very proactive, so an altivec enabled app may be possible at some point. They recently created an optimized Windows app.

Here is a quick comparsion between my iMac G5 1.6 ghz and a Pentium 3.4ghz with HT running Windows. Keep in mind that Rosetta work units vary dramatically in size

With HT on the 3.4 ghz Pentium is effectively broken into two 1.7 ghz processors. The times below therefore reflect the time for a single logical 1.7 ghz processor.

I have also included an AMD Athlon 2400. The windows optimized app favors AMD over Intel.

- iMac G5 1.6 ghz w/1 gig RAM, 5-5.8 hours average.

- Pentium 4 3.4 ghz w 512 mb RAM, 4-5 hours average per work unit with one logical CPU (1.7ghz)

- Athlon 2400 (2.0ghz) w 512mb RAM, 2.5 to 3.5 hours average.

Keep in mind that the above numbers reflect an optimized Windows app against an unoptimized Mac app. The more Mac users on the project, the better the chance they will consider an optimized Mac app. The science benefits from a higher volume of work units completed, so faster is better to them.

Also consider the above numbers very general approximations. Rosetta workunits vary significantly in time from one to the next. On my iMac G5, the shortest work unit was 2.03 hours, and the longest to date is 6.32 hours. The averages above are a 20 work unit sample from each machine.

[99.94]

Thanks very much. That's a pretty reasonable answer. I'll wait a few weeks for some more of the wrinkles to be ironed out, then I will look at it again.

[Shaktai]

Well, some team called "Answer Monkeys" knocked us down a notch in RAC in Rosetta, but only by a little. Now #8 out of 389 teams. We have 9 team members who have returned results.

If we can get just one more person to turn in some results, maybe we can convince reader50 to do stats for the Rosetta team.

It looks like there is a good chance a 10.3.9 app will be available by the end of the week. If so, maybe that will open the door for a couple more folks to join up. We only need 2100 more points per day to eventually move into the #5 spot. Here is hoping this Wednesday brings some sweet new dual core PowerMacs, at an affordable price point. After seeing what Apple managed to pull off with the new iMac, I think there is some hope.

Don't want to pull anyone off of Einstein or Predictor though. Those teams need all the power they have.

[Shaktai]

Just Posted at Rosetta@Home. It should now run on OS 10.3.9. This is good.

October 20, 2005
The rosetta application has been updated on all platforms. The updated versions should have a smaller memory footprint and the OSX version should also run on 10.3.9+.

[Snake_doctor]

Originally Posted by Shaktai

Just Posted at Rosetta@Home. It should now run on OS 10.3.9. This is good.

October 20, 2005
The rosetta application has been updated on all platforms. The updated versions should have a smaller memory footprint and the OSX version should also run on 10.3.9+.

Ok, ok. So when I started with BOINC I joined BOINC Synergy despite being a Mac guy. After all they have nice stats (and they include R@H). Rumor has it that if you change teams at R@H all your credits move with you. Apparently this is the only project where this happens. So I will move my R@H flag to MacNN. Right now I am at number 18 on the Synergy team for that project, and according to them I will not be passed by any of their team members, so I might just be able to help you guys out.

Besides I have worked two Beta projects with Shaktai, and I like his style. So if I do this Mikkyo has to do the BOINC 5.2.5 optimized. Like the "cable guy" says "Get her dunn"

Regards
Phil

[Snake_doctor]

Originally Posted by Snake_doctor

Ok, ok. So when I started with BOINC I joined BOINC Synergy despite being a Mac guy. After all they have nice stats (and they include R@H). Rumor has it that if you change teams at R@H all your credits move with you. Apparently this is the only project where this happens. So I will move my R@H flag to MacNN. Right now I am at number 18 on the Synergy team for that project, and according to them I will not be passed by any of their team members, so I might just be able to help you guys out.

Besides I have worked two Beta projects with Shaktai, and I like his style. So if I do this Mikkyo has to do the BOINC 5.2.5 optimized. Like the "cable guy" says "Get her dunn"

Regards
Phil

Ok so I have done the deed, and as expected along with the first WU I completed after moving to team MacNN, all my credits were posted to the team. So now I will join with Shaktai in requesting some of you guys to come over to R@H and crunch if you have the resources to spare. It is important science, the project team is one of if not the best in support to the users, and as a bonus, MacNN is number 10 in RAC and 9th in total credit today.

I just got my dual G4 running at R@H again yesterday an that should help the team RAC as the effect of that starts to build, but to do that I was forced to migrate that machine to BOINC 5.2.5. Since it is not optimized that fact cuts by 2/3 the credits that machine can accumulate.

MacNN could move up at least one slot soon on total credit but if there were a few more systems it would make a big difference.

Regards
Phil

[ou812nv]

Hello all,

I join this team today, I hope it was ok if not let me know.

[Shaktai]

Originally Posted by ou812nv

Hello all,

I join this team today, I hope it was ok if not let me know.

Of course it is okay. We always welcome newcrunchers. Of course you have to buy a round at the club bar.

Just Kidding, welcome ou812nv. Glad to have you on board.

[Shaktai]

Originally Posted by Snake_doctor

Ok so I have done the deed, and as expected along with the first WU I completed after moving to team MacNN, all my credits were posted to the team. So now I will join with Shaktai in requesting some of you guys to come over to R@H and crunch if you have the resources to spare. It is important science, the project team is one of if not the best in support to the users, and as a bonus, MacNN is number 10 in RAC and 9th in total credit today.
l

Welcome aboard Snake Doctor. Every CPU cycle helps, and I've dropped a few recently, but they were replaced with a very nice Olympus E-500 DSLR and two lenses.

For those who don't know, Rosetta just had their first "big" problem. Thinks got a little sluggish for about 24 hours while they were purging their results database. Not quite a SETI outage.

Might just leave a subtle hint for reader 50, that we now have 14 members that have returned credits. (Rosetta stats pages, hint, hint)

[Snake_doctor]

Originally Posted by Shaktai

Welcome aboard Snake Doctor. Every CPU cycle helps, and I've dropped a few recently, but they were replaced with a very nice Olympus E-500 DSLR and two lenses.

For those who don't know, Rosetta just had their first "big" problem. Thinks got a little sluggish for about 24 hours while they were purging their results database. Not quite a SETI outage.

Might just leave a subtle hint for reader 50, that we now have 14 members that have returned credits. (Rosetta stats pages, hint, hint)

Well I have to plead guilty to being more of a video guy, but my wife likes your new camera.

I have been having some issues with one of my machines as well. I think I have it solved now so my RAC should be on the rise now. MacNN team has moved up one position in the last few days too. So I think since just the 16 of us are holding down the fort, and making progress, the least they can do is add us to the stats. Of course if you really want to drive this home someone could point out that BOINC Synergy list R@H in their stats.

Regards
Phil

[Shaktai]

Originally Posted by Snake_doctor

Well I have to plead guilty to being more of a video guy, but my wife likes your new camera.
......
Regards
Phil

Well here's a couple of links to some of the first photos. I still have to get them organized. Must be time to buy "Aperture".
http://www.pbase.com/brdavid/
http://eaglesnest.smugmug.com/gallery/921432

[Snake_doctor]

Originally Posted by Shaktai

Well here's a couple of links to some of the first photos. I still have to get them organized. Must be time to buy "Aperture".
http://www.pbase.com/brdavid/
http://eaglesnest.smugmug.com/gallery/921432

Nice photos. You shoot for a living or just for fun?

[Shaktai]

Originally Posted by Snake_doctor

Nice photos. You shoot for a living or just for fun?

Just for fun these days. Used to do military, industrial and medical photography as a profession though. Mostly boring, non-creative stuff with a lot of labwork.

[Snake_doctor]

Originally Posted by Shaktai

Just for fun these days. Used to do military, industrial and medical photography as a profession though. Mostly boring, non-creative stuff with a lot of labwork.

Yea. I went the other way. i used to shoot for fun and now it is part of the way I make a living. I do a lot of watching too. Video analysis work. Take a look at A&E city confidential tonight at (8:00 EST), I spent a lot of time working the video evidence on that case.

[Shaktai]

Okay, it sounds like they will soon be releasing the Rosetta code for optimization. Posted today from David Baker.

We will be distributing the code really soon--it would be great if people could try Bill's suggestion out and let us all know how it goes!

Bill's suggestion was for people to visit the MacNN forums to find out how the SETI client was optimized.
We're infamous.. er I mean famous folks.

Hopefully we can find some individuals willing to give Mac optimization a shot. David Kim had looked into it, but the priority was to get the Mac app compatible with 10.3.9. Apparently he isn't real familiar with Mac compiling so is learning as he goes. It was either optimize for 10.4.x and not have 10.3.x users participate, or compile for compatibility with 10.3.9 and not have altivec.

[OneMacGuy]

I am now signed up and working on Rosetta@home with 3 CPUs. My Dual G4/533 is happiliy chugging away and I have a 2.66 P4 working on Rosetta also. I hope that helps the team some. If I like where this goes I might be able to move some serious power over here.

[Shaktai]

Originally Posted by OneMacGuy

I am now signed up and working on Rosetta@home with 3 CPUs. My Dual G4/533 is happiliy chugging away and I have a 2.66 P4 working on Rosetta also. I hope that helps the team some. If I like where this goes I might be able to move some serious power over here.

Very Cool, and we know what kind of serious power you have. You can knock me out of the number one spot any day you like. It's a good project that looks like it can have a real impact.

[OneMacGuy]

Originally Posted by Shaktai

Very Cool, and we know what kind of serious power you have. You can knock me out of the number one spot any day you like. It's a good project that looks like it can have a real impact.

I am having a strange problem woth the Boinc Manager on my old G4. I had to quit and restart the app today and whenever I run the Boinc Manager now, it says Connection Error, the password you have provided is incorrect, please try again. My password has not changed, I even went and logged onto the Rosetta web site and logged in there to make sure someone had not changed it for me. Any ideas?

[Shaktai]

Originally Posted by OneMacGuy

I am having a strange problem woth the Boinc Manager on my old G4. I had to quit and restart the app today and whenever I run the Boinc Manager now, it says Connection Error, the password you have provided is incorrect, please try again. My password has not changed, I even went and logged onto the Rosetta web site and logged in there to make sure someone had not changed it for me. Any ideas?

Yep, it is an irregular problem we see with the Mac version of BOINC manager. For some reason you have to restart the computer to get it to connect again (with autostart at launch). It will actually crunch in the background but no upload or download. I get the same problem. Try not to quit. Haven't tried the new 5.x BOINC version yet, so don't know if they fixed it.

[OneMacGuy]

Originally Posted by Shaktai

Yep, it is an irregular problem we see with the Mac version of BOINC manager. For some reason you have to restart the computer to get it to connect again (with autostart at launch). It will actually crunch in the background but no upload or download. I get the same problem. Try not to quit. Haven't tried the new 5.x BOINC version yet, so don't know if they fixed it.

Yep, thanks! That worked.

Puerto Rico?

[La Muis]

I 'm running R@H on a dual G5-2Mhz with the 4.44 superbench Cli Client. It ran fine until a few days ago. Now i'm only producing 'max cpu time exceeded' errors with lots of 'no heartbeat from core client - exiting' messages.
It seems i'm getting bigger wu's lately from R@H and there is going something wrong with the superbench client when, in my case, the cpu time is more then 22k seconds.
When i switch back to the standard 5.2.5. client i can finish large wu's without any errors.

I looked in some of your results and see the same thing is happening there. Someone any ideas?

(My powerbook G4 doesn't seems to have any problems with the superbench client. But it isn't running 24/7 so mabye it is still crunching 'old' wu's).

Me too. Have you posted anything over at Rosetta@home forums on their site?

I'm running BOINC Superbench menubar 4.44 on a G5 dual, and notice that my Rosetta crunches are getting errors since Nov 19.

[Shaktai]

Originally Posted by OneMacGuy

Puerto Rico?

It's a family thing. Recently the last of my mother's brother's and sister's passed away, including the patriarch for both our families. I crunch for Puerto Rico in their memory. That was where my mother was from, and most of her family including most of my cousins (primos) still live. I live in Colorado though.

[Shaktai]

Originally Posted by Andrew F

Me too. Have you posted anything over at Rosetta@home forums on their site?

I'm running BOINC Superbench menubar 4.44 on a G5 dual, and notice that my Rosetta crunches are getting errors since Nov 19.

I just looked at mine and experienced the same thing on two out of three work units. The two that errored were "big" units. Haven't had a chance to check my PC's or the Rosetta forums yet.

Seems to be a particular type of work unit: 1hz6A_abrelaxmode_random_length10_jitter02_omega_s im_aneal_bab90_.....

Seems to primarily impact Macs for me, but based upon comments in the forums, looks like it impacts other machines as well. I'm watching one now on my G5, and if it errors, I am going to abort the remainder of these work units in queue

[Shaktai]

Originally Posted by Shaktai

Seems to be a particular type of work unit: 1hz6A_abrelaxmode_random_length10_jitter02_omega_s im_aneal_bab90_.....

Well, just looked at my last unit, and it completed fine. Long one though, took about 10.5 hours. Maybe it was just a temporary work unit glitch. I'll keep watching the next three in queue.

[OneMacGuy]

Originally Posted by Shaktai

Well, just looked at my last unit, and it completed fine. Long one though, took about 10.5 hours. Maybe it was just a temporary work unit glitch. I'll keep watching the next three in queue.

Is there a G4 optimized version of the Boinc client that will work with Rosetta?

BTW, I am now number 2 in Rosetta for Team MacNN! I only have to increase my production by at least 8 times to catch Shaktai!

[Shaktai]

Originally Posted by OneMacGuy

Is there a G4 optimized version of the Boinc client that will work with Rosetta?

BTW, I am now number 2 in Rosetta for Team MacNN! I only have to increase my production by at least 8 times to catch Shaktai!

Optimized BOINC clients? Yes, aif your are using BOINC 4.x then any of the following will work. http://members.dslextreme.com/~reade...eam/boinc.html Mind you they only improve benchmarks (and credits), they don't improve actual speed. There is no Altivec optimized Rosetta apps yet. The source code isn't public, but they have been talking about it.

Catching me wouldn't be that hard. My production is decreasing, not increasing. Increasing by 6 times should do the trick.

Keep forgetting to buy that winning lottery ticket so I can get a dozen quad PowerPCs.

All my Rosetta WUs have been erroring out for the last five days. There's some speculation that the 4.x BOINC doesn't work on some machines with Rosetta, and that upgrading to BOINC 5.25 is recommended. Since I prefer the Altiveced SETI and Einstein on the superbenched 4.44, I'll be waiting detaching from Rosetta until BOINC gets a superbenched 5.25 -- any word from Mikkyo on that?

[OneMacGuy]

Originally Posted by Shaktai

Catching me wouldn't be that hard. My production is decreasing, not increasing. Increasing by 6 times should do the trick.

Keep forgetting to buy that winning lottery ticket so I can get a dozen quad PowerPCs.

Better get that lottery ticket... I am up to over half of your rate now and I may move some more of my home farm over to Rosetta over the holidays. I have been really amazed by how much more production I get out of Athlon XP chips vs. Intel P4 boxes. I have an older Athlon XP2500 Barton based system that is smoking a 2.8 P4 Prescott with HT. It is actually doing more than one 2.8 P4 running HT and one that is not, not much difference between those two. My old Dual 533 G4 system is really lagging though. This really makes it show it's age. It is running neck and neck with my wifes 1 Ghz G4 emac though.

[Shaktai]

Originally Posted by OneMacGuy

Better get that lottery ticket... I am up to over half of your rate now and I may move some more of my home farm over to Rosetta over the holidays. I have been really amazed by how much more production I get out of Athlon XP chips vs. Intel P4 boxes. I have an older Athlon XP2500 Barton based system that is smoking a 2.8 P4 Prescott with HT. It is actually doing more than one 2.8 P4 running HT and one that is not, not much difference between those two. My old Dual 533 G4 system is really lagging though. This really makes it show it's age. It is running neck and neck with my wifes 1 Ghz G4 emac though.

Yeah, I noticed you are nearly on my tail. Think I've got enough of a lead though to hold you off for a month or two.

[Shaktai]

Nice power-up by onemacguy and jeffreyloaf. You guys aren't thinking about trying to get past me are you?

[crunch]

Anyone else notice thier stats not being credited or updated for the past 10 days or so?

BTW Shaktai, One Mac Guy has you in thier sights hee hee

[Shaktai]

Originally Posted by crunch

Anyone else notice thier stats not being credited or updated for the past 10 days or so?

BTW Shaktai, One Mac Guy has you in thier sights hee hee

Yea, I noticed OneMacGuy trying to creep up behind me. Stats are being credited as near as I can tell though.

[crunch]

Hmmm, Einstien's been updating normally, but nothing's been credited for me in the past 11 days now for Rosetta.
If they'd turn on the "Pending Credit" part of the website I'd be able to whine at them *lol*

Thanx, was wondering if it was the entire project. Guess it's not.

[Shaktai]

Originally Posted by crunch

Hmmm, Einstien's been updating normally, but nothing's been credited for me in the past 11 days now for Rosetta.
If they'd turn on the "Pending Credit" part of the website I'd be able to whine at them *lol*

Thanx, was wondering if it was the entire project. Guess it's not.

Check your computer list. Rosetta doesn't use pending credit, because it doesn't require the validation process of other projects. Make certain you aren't getting a lot of work unit errors. The work unit is either good or not, and if good you get credit immediatly. Recent Rosetta work units have been much longer as well. I use at least a 2-3 hour cycle between switches if running Rosetta with other projects, to ensure it can get to the next "step" before switching. On slower machines if you run multiple projects it may keep resetting to the last "step" and not have time to finish if you switch projects every hour.

[Snake_doctor]

Originally Posted by Shaktai

Check your computer list. Rosetta doesn't use pending credit, because it doesn't require the validation process of other projects. Make certain you aren't getting a lot of work unit errors. The work unit is either good or not, and if good you get credit immediatly. Recent Rosetta work units have been much longer as well. I use at least a 2-3 hour cycle between switches if running Rosetta with other projects, to ensure it can get to the next "step" before switching. On slower machines if you run multiple projects it may keep resetting to the last "step" and not have time to finish if you switch projects every hour.

Shaktai,

I think I have stumbled on a problem that seems to be aggravated by the optimized Client and was wondering if you have been seeing the same thing on your Macs. It seems as though if you get a very short WU followed by a very long one, the long one will error for taking too long to finish. The WU runs normally in every respect but as the system adjusts the estimated time to completion for the short WU it does not provide enough time for the next one if it is too long. Obviously the optimized client does not help because the system expects the WU to complete faster based on the benchmarks. It looks as though so long as the WUs are all within about 30% of each other on completion times it will run ok.

So far I have only seen this on R@H, and part of the problem is that they have implemented a fix for the 1% stall problem that will error a WU if it takes too long. They really should check both the CPU time and the Percent complete to determine if a WU has stalled or not, but right now all they look at is the CPU time.

Have you seen this issue?

Regards
Phil

PS- Any suggestions on projects where a few more MACNN folks could make a difference? Also thanks to all who worked on the optimized client.

[Shaktai]

Originally Posted by Snake_doctor

Shaktai,

I think I have stumbled on a problem that seems to be aggravated by the optimized Client and was wondering if you have been seeing the same thing on your Macs. It seems as though if you get a very short WU followed by a very long one, the long one will error for taking too long to finish. The WU runs normally in every respect but as the system adjusts the estimated time to completion for the short WU it does not provide enough time for the next one if it is too long. Obviously the optimized client does not help because the system expects the WU to complete faster based on the benchmarks. It looks as though so long as the WUs are all within about 30% of each other on completion times it will run ok.

So far I have only seen this on R@H, and part of the problem is that they have implemented a fix for the 1% stall problem that will error a WU if it takes too long. They really should check both the CPU time and the Percent complete to determine if a WU has stalled or not, but right now all they look at is the CPU time.

Have you seen this issue?

Regards
Phil

I have seen a higher error rate using the superbench client and that is probably one of the reasons why. On Rosetta, the superbench is pretty much considered "cheating" because there is no tempering process of comparing against other units for validtion. On the beta page, there is now a "non-superbench" version that improves benchmarks a little for equity on Rosetta with the PC's but is significantly less then the superbench. It also seems to have fewer errors, which makes me think your theory above may be correct. What errors I do get are those that error immediatly, instead of after 20-30,000 seconds. (they error on all machines, not just Mac so are just bad units).

[Snake_doctor]

Originally Posted by Shaktai

I have seen a higher error rate using the superbench client and that is probably one of the reasons why. On Rosetta, the superbench is pretty much considered "cheating" because there is no tempering process of comparing against other units for validtion. On the beta page, there is now a "non-superbench" version that improves benchmarks a little for equity on Rosetta with the PC's but is significantly less then the superbench. It also seems to have fewer errors, which makes me think your theory above may be correct. What errors I do get are those that error immediatly, instead of after 20-30,000 seconds. (they error on all machines, not just Mac so are just bad units).

I might give the non-superbench version a try. The superbench really jacks up the claimed credit. Actually a little too much for my taste. I don't mind an ajustment to compensate for inequity but four to five time the normal credit is a little over the top.

Regards
Phil

[reader50]

[Shaktai]

Thanks reader50.