Joined: 16 May 22
I'm on an 8C16T cpu.
I keep BOINC running at 25% or 50% at times.
When I first added this project, it began running 4CPU tasks, 1 at a time. When a task finishes there is idle time until the next one starts.
At the same time it will not allow any other project to run. Several of the projects I run have stalled (hence me adding this one), but one is active, and Milky way stopped all progress on it. Resource allocation is equal among all projects. So 50/50 in this case.
There was also another recent thread that suggested creating an app_config.xml file.
I created the following file:
<app_config> <app> <name>milkyway_nbody</name> <max_concurrent>2</max_concurrent> </app> <app_version> <app_name>milkyway_nbody</app_name> <plan_class>mt</plan_class> <avg_ncpus>1</avg_ncpus> <cmdline>--nthreads 2</cmdline> </app_version> </app_config>
with the intent of allowing it to run a couple of tasks concurrently, to eliminate the idle time I was seeing.
This config pegged my CPU at almost 100%, disregarding boinc's usage limits.
Then I deleted the app_config, and now Milkway is no longer doing nbody WUs, but it is doing multiple seperation WUs.
And still choking out other projects.
Joined: 8 May 09
I'm on an 8C16T cpu.
First you are getting Separation units because you haven't unchecked the box in your preferences, if you uncheck the box and then save your changes they should stop coming.
As far as your app_config file I use a simpler one:
Edit the <max_concurrent> line to suit your needs and the <nthreads line as well>, for me it means no more than 6 tasks at a time and to only use 2 cpu cores per task. By default MilkyWay NBody tasks will use all available cpu cores you let Boinc use when you ask for new tasks. That has led some people to use a zero resource share project and to manually change the cpu cores available to Boinc everytime they get new MilkyWay NBody tasks, to me that's a pain so I use the app_config file instead.
As for Boinc not running gpu tasks steadily MilkyWay has a 10 minute time out from the time you return a task to when you get a new task, this is only for gpu tasks cpu tasks are not affected. A prior Admin made the change and the current Admins can't seem to find the setting to change it. An easy work around is to again have a zero resource share gpu project pick up the slack for those 10 minutes so MilkyWay doesn't keep asking for tasks and getting into the waiting loop.
©2023 Astroinformatics Group