First off - yup me also a Pentathlon Rosetta Newb. I came in Monday night to have a quiet spin round, get the feel of Rosetta, try some short WUs {more of that later } ..... and promptly hit the buffers in time honoured newb fashion :) Mod.Sense very kindly picked me up, pointed out a few home truths, and got me on the right track. It was suggested I start a thread for Pentathlon Rosetta Newbs on Q&A best practice etc. So here we are.

Scratched the head a little, and thought, yup a central place to quickly get advice would be neat for us Pentathlon Visitors. So here it is ... To start it I thought I would briefly relate the advice I was given, and pointed at, that would immediately affect an enquiring Pentathlete who's sole aim in life is get up and running for their Team with minimum hassle:)

- Select WU size in Preferences, select 3 hours or above, smaller size will cause issues, not worth it for us, summary of issue:

... expectations of what a 1hr work unit is are not realistic for R@h..... when you do that, some of the other nice things such as accurate progress %, and consistently completing within such a limited timeframe go out the door. Each task must complete at least one model. For some tasks you will see a model every 5 minutes or so, for others, it can take several hours. So, not all tasks are going to complete within your one hour target, and that is normal and to be expected.

- It would seem that BOINC gets more confused than it usually does if frequent changes are made to WU length, the advice is apparently settle on a size of 3-4 hours or above, and stick with it.

- Progress Percent not advancing?

A: Rosetta recomputes the progress percent at the end of each model. The model number is shown in the graphic. As long as the "steps" are continuing to progress, it is working. Once it completes the model it is working on, it will recompute the progress. At that point the progress % will be determine by looking at the time it took to complete the first model, as compared to your WU runtime preference. If your WU runtime preference is low (<4hrs) you will frequently see the progress % jump from 1.xx to 100%, or into the 50% range. Basically, each different protein takes a different time to crunch a model. Some proteins will crunch for several hours to complete a single model. Others will crunch a model every 5 or 10 minutes. It is the nature of the science being done with Rosetta.

BOINC and Rosetta have measures in place that will abort work units that aren't running properly. So, in general, unless you see some specific advice to the contrary, you should NOT abort work units (WUs). {sic: no prizes for guessing what I did ...}

- "To completion" time is going UP!
A: This is normal. The time gets revised when the progress % is recomputed at the end of each model. So at the end of a model the time to completion will drop. Then during the crunching of the next model it will increase again.

- I'm familiar with SETI and BOINC already, but what should I know about Rosetta? See this link:

That'll do .... I think ..... and will get most up and running, any more becomes a War and Peace novel for Pentathelon Visitors, not what the thread is for.

Questions to the local experts - and that decidedly is not me! - by posting on the thread.

Local Experts, any susinct hints of best practice for the Pentathletes would doubtless be greatly received, and may result in some Rosetta converts. Your words of wisdom are sought after :)

Regards
Zy

This thread is definitely a good idea! Hope those taking part have fun and hopefully some will stick around to crunch Rosie afterwards too ;)

To expand on how the Rosetta tasks relate to models: Rosetta will run as many models within a task as it can, given the time preference that you can set under Participants > "Resource share and graphics: Rosetta@home preferences" here on the web site when you're logged in. There's no need to change these settings though, unless you have limited band-width (e.g. dial-up), in which case a longer duration might help.

Regardless of the task duration, a faster computer will do more models. A slow computer might do 5 models within a given task in a 4-hour task-length, whereas a faster computer might do 10 models in that same task. The computer that does 10 models will get twice the granted credit of the computer that does 5 models, regardless of the claimed credit.

Claimed credit is largely irrelevant (because it's based on the overly-simplistic BOINC benchmark), so don't worry if your claimed and granted credit are completely and consistently different - granted is a more appropriate measure of how well your machine crunches Rosetta tasks.

Hope that helps and doesn't confuse things! Basically, Rosetta should be set-and-run and faster computers will get more credits than slower ones. :D

---

Hope I've not committed another faux par ... :)

Learning from yesterday, I made sure the cache was run dry after a good overnight/morning session of 3 hour Models, then set the machine to my "work" preference where I had set it for 12 hours - aim being to load up 3-4 days worth of 12hr preference models for six cores on a 1090T.

Down came the models. So far so good. They then proceeded to error out one by one within 30 seconds to 3 mins of download ..... all of them. I guess I did something wrong, but cant fathom what - I ran it dry, changed to the 12hr preference, waited 10/15 seconds then started the download. Changed nothing else from the previous 12/15 hr successful trial crunching session.

Enlightenment greatfully received ... I dont want to set off another download batch and trash those. I managed to catch the last third or so and abort transfer to stop models trashing.

EDIT: Set cache this end to zero to limit and trashing, until find out whats happening. Downloaded one with preference set to Work (12hr), trashed within 2 mins. Going to set it back to default preference, see what happens.

Regards
Zy

Hope I've not committed another faux par ... :)

Learning from yesterday, I made sure the cache was run dry after a good overnight/morning session of 3 hour Models, then set the machine to my "work" preference where I had set it for 12 hours - aim being to load up 3-4 days worth of 12hr preference models for six cores on a 1090T.

Down came the models. So far so good. They then proceeded to error out one by one within 30 seconds to 3 mins of download ..... all of them. I guess I did something wrong, but cant fathom what - I ran it dry, changed to the 12hr preference, waited 10/15 seconds then started the download. Changed nothing else from the previous 12/15 hr successful trial crunching session.

Enlightenment greatfully received ... I dont want to set off another download batch and trash those. I managed to catch the last third or so and abort transfer to stop models trashing.

EDIT: Set cache this end to zero to limit and trashing, until find out whats happening. Downloaded one with preference set to Work (12hr), trashed within 2 mins. Going to set it back to default preference, see what happens.

Regards
Zy

All the tasks are the same with regards to time required regardless of your duration preference - Rosetta will just fit more models into a given task if you increase the run time. Whereas a task in most projects is quite straight forward, here each one can contain multiple models. Each model is based on the same starting information but with slightly different parameters each time. I don't think there's much point in changing the run time though- it won't make any difference to your credit or throughput. Also, no need to empty the queue if changing the run-time because rosetta will just keep running more models using the existing tasks, but rosetta might not notice the change without a boinc restart or until it starts a new task.

---