Quote Originally Posted by trigggl View Post
I have the following suggestion for naming threads in the app_info.xml project forums. Hopefully, this will provide clarity when looking for an app_info for a cruncher's particular needs. If there are no objections we'll need administrative help to rename the ones already posted.

Also, it would probably be a good idea to have a guideline sticky detailing the convention for new threads that users will post. Since it would probably clutter up the forums by discussing app info availability and other app_info topics that don't provide an app_info we should probably leave/move those to the top level forum (sorry for breaking this with my Moo! app_info question). This will eliminate the need for stickys in each individual Project app_info forum. This can also be included in the guideline sticky.

Here is my proposal:

Code:
<Project> [app-]<version> (<plan_class>) <arch>/<os> - <nthreads> WU
  • project_______SETI, Milkyway (MW), Moo! Wrapper (Moo!), etc
  • app__________a project may have more than one app
    _____________optional if there's only one app at the project
  • version_______the version of the app
  • plan_class____ati14, cuda , opencl_ati_100, etc
  • arch_________i686, x86_64, ppc64, etc
  • os___________win, linux, mac, solaris, etc
  • nthreads WU__1, 3, 8, number of tasks at a time


So, using Poem as an example:

Project is Poem. Only one GPU app at the moment, so not specified. The current app_info in the Poem forum is 0.1, so version is 0.1. The plan class for this app_info is opencl, so currently opencl_ati_100. Arch is x86 which equates to i686 in the boinc client. OS for this one is Windows (win). The tasks at a time is 8.
Sounds good. Not sure the nthreads part is needed since, depending upon the project and the hardware being used, and whether it is for dedicated crunching or not will make a difference. Most app info files are for running a single WU at a time and adjusting the command line parameters. Some affect how many WUs to run at a time, but those (think POEM) also require messing around with both the avg_cpus as well as the count of the GPUs. So, rather than include threads in the subject, my suggestion would be to have a section for notes that explains how to tweak the app info (e.g. to run 6 POEM WUs on 1 GPU , you set the avg_cpus to 1 but set the count to 0.166 which is 1/6; to run 4 WUs on 1 GPU allocating 50% of a CPU for each WU, use avg_cpus = 0.5 and count = 0.25; etc ). That way those who are long division challenged can look up the settings and we wouldn't need to have 25 versions of the exact same app info. And, when you figure one per platform per OS type per GPU type, that could be hundreds of the same one with only the counts being different.