I am revisiting this question because I have one system where I have been able to keep World Community Grid from flooding the running tasks to the point where nothing else would run and using the same general ideas I haven't quite been able to manage it on another system.

I am running two systems that run multiple projects and one system that is 100% Seti, except when it runs out of gpu tasks, then it downloads some gpu tasks from Einstein at Home.

This system http://setiathome.berkeley.edu/show_host_detail.php?hostid=8281049 is your basic 4c/8t system with a Gtx 750Ti (at the moment) running it. It is currently using Win10.

Setiathome (1000.0)
SetiathomeBeta (1.0)
EinsteinatHome (gpu stand by) (0.0)
WorldCommunityGrid (1.0)
RosettaatHome (2.0)
MindModeling@Beta (1.0) (no tasks usually)

I have used the <project_max_concurrent>1</project_max_concurrent> in the individual project folders to control how much access WCG and Rosetta has to the cpu cores.

So far, the scheduler has not refused to download Seti tasks because there were too many WCG and/or Rosetta tasks.

However, I have repeatedly had to go to No New Tasks and sometimes abort WCG/Rosetta tasks on this system: http://setiathome.berkeley.edu/show_host_detail.php?hostid=8565979 It is a 8c/16t system running a Gtx 1060 3GB under Linux/CUDA91.

It is also running RNA World otherwise, the list and resources settings are the same as the first computer.

I have used the <project_max_concurrent>4</project_max_concurrent> in the individual project folders to control how much access WCG and Rosetta has to the cpu cores.

The reason I am certain that the scheduler is not keeping the Seti cpu queue full is I run into cases where the only tasks running on the cpu are WCG and Rosetta. So basically I have 6 threads not running at all (I run 2 cpus for my gpu) when the WCG/Rosetta are crunching along as hard as I will let them.

The event log seems to be reporting something like no new tasks needed, computer has all it can handle. And it has no seti CPU tasks.

I am not running competing GPU projects so it isn't (I hope) going to be an issue.

I know there is no "formula" here. I am looking for ideas. I am pretty sure there are no parameters for controlling the total number of tasks a specific project can download into the cache. Or is there?

Thank you,
Tom

---

A proud member of the OFA (Old Farts Association).

The "resource share" setting is based on a long term (weeks) work achieved figure, and is not easy to get stable - the figure you are using are "sensible" if you want SETI to have a large share of your effort, and the others only a small amount.
BUT:
BOINC will calculate how many days work you have in hand (although it does appear to ignore the very long runtime CPDN tasks) and will load the cache to that limit. While in theory each project will only deliver tasks to the limit set by your computer some projects are "less good" at doing so and will deliver a large number of short deadline tasks, so swamping others out. Aborting tasks can actually lead to you getting more from that project, which is counter productive.
The best you can hope for is to let the computer run for several weeks, just monitoring it, as, in theory over that time BOINC and the project servers should learn what is possible.

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?