In the performance part of the account page, what exactly does the 'Target CPU time' do? Does it allow your computer to allocate workunits that will take approximately that long? or is that the amount of time Boinc will switch between projects if you are joined to a few projects? Any help would be appreciated.

Thanks

not sure where it says 'performace'?

If you mean the 'target' CPU run time then its the time that rosetta will run for before reporting the task as complete. Each task consists of as many models (decoys) as will fit into that target run time- but as each decoy takes an unknown amount of time it has to be estimated, hence the variable run times. Therefore, on a given computer, running a given protein model, an 8hr task will contain approximately double the number of decoys that a 4hr task will contain.

HTH
Danny

---

The target CPU time doesn't affect what WU you get. Every time the Rosetta app finishes a model, it compares the current run time to the target run time. Rosetta starts another model if there seems to be enough time left.

If you change your target CPU time, then all WUs currently on your computer will use the new target (once the change in preferences propagates to your computer). This includes the WU currently being crunched.

The BOINC client doesn't know about the target run time. The estimate that it gives for how long a WU will take is based on how long WUs took in the past. Once a few WUs have been crunched with a new target time, the BOINC client will learn the new time that WUs are taking and will then adjust its estimate of how long WUs will take to crunch.

I have set the CPU target time to 8 hrs and after 3 hrs the graphs won't display anymore. Is this normal?

---

I have set the CPU target time to 8 hrs and after 3 hrs the graphs won't display anymore. Is this normal?

nope, but i dont think it has to do with target run time, you'de better post it in the problems with version... topic

I have set the CPU target time to 8 hrs and after 3 hrs the graphs won't display anymore. Is this normal?

nope, but i dont think it has to do with target run time, you'de better post it in the problems with version... topic

For some reason it started working just a moment ago.

---

I have set the CPU target time to 8 hrs and after 3 hrs the graphs won't display anymore. Is this normal?

nope, but i dont think it has to do with target run time, you'de better post it in the problems with version... topic

For some reason it started working just a moment ago.

when it happens again, post it :). so they can find out if this is common. could also be your pc hadn't have enaugh cpu to display.

I'm just getting my head around 'target CPU time' and how it works.

a few days ago I set up a new host (id #678950) using an ancient win98 600MHz athlon and it's chugging along fine even if more like 90% winter heater and only 10% number cruncher. I wondered why the work units were behaving differently to my 3.4Gig machine until I realised that the time was similar and hence the work done in that time was less.

it's set on the default 3hrs target CPU time and on analysing the first ten WUs it's averaged 2.5hrs per WU, each earning 5.35 creds and finding 4.3 decoys on av.

in a significant proportion of tasks, because of the limited power, it's only going to find one or two decoys in the three hours. as a result, tuning the target time might optimise the work done. I'm trying to think whether it is better to shorten the time and end the WUs quicker that aren't going to be very efficient at finding decoys or else lengthen and let the easier tasks have a good run and find more of them. there's probably not much in it either way - thanks anyway for thoughts on the matter. /pg

Glaesum, For you use, I've put together all the work for both machines onto one spreadsheet. There are columns for C/H (claimed credit/hour), G/H (granted credit/hour), number of decoys, Seconds/decoy, Claimed credit/decoy, and the WU name. The top is from your P4, the bottom from your Athlon. Both are sorted by WU name for easier comparison.

tony

Basically, think of each task like a jig saw puzzle with each being of differing difficulties. Each person who contributes to the work has differing abilities as well, so you might be able to find and place 5 pcs in one hour, I might be able to do only two/hour. Now, If I work for 3 hours, I'll find and place 6 pieces. If you only do one hour, then you'll just find and place five pieces. I'm not aware of any setting yielding more efficient work (except for the small start up stuff, making longer times minimally more efficient)

Your slower machine is not going to do any better or worse with longer runtimes. A larger slow model will take considerable time, but then... it does on any system. Your is no worse at those then it is good at short models.

The runtime preference is there to help you schedule getting work done on time (before the current 10 day deadline). And if the thing is on all the time, then generally longer runtimes means less network activity to the servers. Thus keeping everything more efficient for everyone. For a user that only has BOINC running a few hours a day, it might take them longer then the 10 days to complete a 24 hour task. Others, it just takes 24 hrs.

Just make any changes to runtime very gradually, over the course of a week or so. This avoids getting too much work, because it gives BOINC some time to learn that the tasks are taking longer, and that it shouldn't request as many tasks to keep your reserve of work at the level you prefer.

---

Rosetta Moderator: Mod.Sense

thanks to both of you, especially for all that work in setting up that spreadsheet. actually I'd copied the first ten wus into a table too but I only calculated the averages for an overview. gosh, did you hand cast each wu name into the sheet - as my global c&p from the web page only picked up the wu numbers? beyond the call of duty I'm sure. :-)

there were two things I was pondering:

1] watching the graphics, the folding seems to go in phases; so does the first phase get repeated with every decoy attempt? if not, then a longer time setting will be better

2] imagine a wu that takes 5800s to find a decoy and then ends; this releases the remaining 5000s of the 3hrs to the next randomly but on average more difficult wu.
shorter settings will mean the proportion of total crunching time that is released and carried forward will increase (ie greater lumpiness which is otherwise smoothed out with either more powerful machines or longer task times)

/Pete

gosh, did you hand cast each wu name into the sheet - as my global c&p from the web page only picked up the wu numbers? beyond the call of duty I'm sure. :-)

Fred W, and I are working on a macro that will get the data I showed if a person has MSexcel. It's not done yet, so yes, I manually collected the WU Name, but the rest is already in the script. I should be "complete" (if anything really ever is) later today. It will make analysis of many things easier.

I'll defer the rest of your query to those more familiar with Rosetta itself.

glaesum, the first phase always occurs. Sometimes, if the energy level after the first phase is very poor, the second phase is skipped and the process begins the next model.

Yes, it is true that, on average you will get to 1/2 of your average model crunch time away from your preferred time, and therefore a faster CPU will tend to complete more closely to the same time for each task, it doesn't matter.

Again, it is how you use your machine. If ending several tasks after 2hrs rather then taking them to the full 3 will not cause you any problems in keeping enough work on hand, it really doesn't matter.

When you start a new model, you begin a new journey. The only relationship to the prior model is that the journey is in the same wilderness, but the point at which you start is completely different, so you can't even TELL that it is the same wilderness, because you begin on a different slope of a different mountain, facing a different direction.

---

Rosetta Moderator: Mod.Sense

very evocative metaphor indeed! :-)

with winter approaching, I wish my knees were up to skiing in those mountains again... {wistful sigh}

Fred W, and I are working on a macro that will get the data I showed if a person has MSexcel. It's not done yet, so yes, I manually collected the WU Name, but the rest is already in the script. I should be "complete" (if anything really ever is) later today.

Oops, hit a snag, works great as long as there isn't an invalid result(I don't have any invalid results, and only found this while using the top 20 hosts as a test), then I get a run time error. This will be delayed a bit longer while debug continues.

very evocative metaphor indeed! :-)

with winter approaching, I wish my knees were up to skiing in those mountains again... {wistful sigh}

I was PICTURING being "lost in the wilderness" when I wrote it, and all trees look a lot alike... but wanting to ski a different slope every time down the hill, and getting to the bottom only to realize you had already done that one certainly still applies :)

---

Rosetta Moderator: Mod.Sense