Posts by dream_it_true

1) Message boards : Cafe Rosetta : Moderator contact thread archive (Message 31715)
Posted 27 Nov 2006 by dream_it_true
Post:
The only thing that I've found that seems to fix this is to go to

"rosetta@home"

then

"your account"
"preferences"
"rosetta@home preferences" (not the general preferences)

In this area your "resource share" is probably set to 100 (the most likely setting you might pick from when you set up the project- I'm not certain what it defaults to if you don't pick a value, though 100 would make sense). The documentation is kind of vague and it took me while to figure this out (right or wrong). I don't think this should be thought of as 100%: It is more of a priority (kind of like mainframe priorities or Unix nice-ing) - at least as I understand it. It's value might be 100, but that does not mean 100% of resources. Rather, you add the various projects' priorities together - whatever they are - then divide the sum, by the number of projects, for the mean. Projects above this value/point get timeslice priority above "average" or actually above lower projects regardless of the average. Essentially they seem to be asked "do you need attention" more often than the other projects. Note: If you use a default (or the same) value for everything the math will obviously work out to that value. This value might be equal across all projects at 100 (or whatever) but the vaguaries of the way the projects operate within BOINC evidently does not guarantee that they really get equal "functionality" out of BOINC. I WOULDN'T ACTUALLY DO THE MATH AT ALL - NO NEED- IF THE NUMBER IS HIGHER THE PROJECT GETS MORE ATTENTION. I don't think it works out lineal so no reasn to do the math. Just set Rosetta (or which ever project you are not getting response from) higher than the other projects you are running. To do this

click on the bottom at

"edit Rosetta@home prefernces"

edit the "resource share" field's value and input something like 150 or 200. It will take a few minutes to reflect the update, but it will then preferentially prioritize Rosetta. It still runs all the other projects you may have running for their full time slices - which is a different setting - but will now give Rosetta the priority it needs to get new work. It will come back to Rosetta more often so rosetta may end up getting more done (at the expense of others). If this happens you can always scale the Rosetta preference slowly back to as close to 100 as will permit new work to still come down.

In particular, I've noticed if you leave the default settings of 100 for everything, on some machines Rosetta has a hard time contending with Einstein@home. This completely fixed my allocation and timeslicing problems. I now run with Rossetta at 150 and Einstein and Seti both at 100. Eistein still completes more work than Rosetta does, but Rosetta always runs and gets work now.

Note: This does not result in an excessive "slew" or "swap-out" condition with wasted time unloading from memory. That evidently is a different setting. Each of my projects runns for their full 3 hours slice (I upped that too from the default 60? minutes) before swapping out.

I've tried to find other solutions before finding this and the only other thing I can think of is to make the

"target CPU time"

setting in a given project closer to the timeslice or the number of times a day it checks for new work -but I'm completely speculating there - this fixed my probelm so I haven't looked further.

hope this helps,

Jeff A


*********************** In response to ***************************
I don't know if this is the right place to post this, but.....

Something seems to be amiss on my end. I haven't gotten any new work units for a couple of weeks, and am wondering what went wrong.

I have tried resetting, have detachedd and re-attached to the project but these actions didn't seem to help. Any suggestions?[/quote]

Are you BOINCing any other projects? If you have more than two projects on a machine[1], it may be simply that those projects "beats" Rosetta to the "get next project".

I've found mixing two projects on each machine will work better than having three or four projects on one machine.

. . .

Mats






©2024 University of Washington
https://www.bakerlab.org