Albert and Einstein work request effects

Joined: 6 Dec 05
Posts: 10
Credit: 67,924
RAC: 0
Topic 84863

I'm quite new to Albert, though I've been around on Einstein for years.

The two hosts on which I've enabled Albert are puzzling me in work-fetch behavior. While I have the Albert resource share set to 2 and the Einstein share set to 149, they seem to be fetching a lot of Albert work and grinding their Einstein queues down to very low levels. For example, one host currently has eleven Albert CUDA WUs and just one Einstein CUDA WU, and the other currently has eight Albert CUDA WUs, and five Einstein (three executing and two in queue).

Is there an intended or observed cross-coupling between Einstein and Albert work fetch, or am I more likely just another meddling user puzzled by the vagaries of the scheduler, especially in fresh project startup situations? Both are running the 7.0.28 version of BOINC which I think remains the current fully released version.