I noticed today that there's a new GW tuning run. O1Spot1THi is extremely memory-intensive, and it was only running on three of the allowed four processors (4 physical, 8 hyperthreaded) because it hit BOINC's memory limits. The fourth instance was suspended, waiting for memory to open up.
On my machine, each active instance is consuming 1.75GB of swap and anywhere from 0.82-1.75GB of RAM.
I thought I should mention this because if these tasks consume this much memory normally, they'll hit memory limits on other machines as well. This will lead to faster than normal times on each workunit, but slower progress overall due to the unused CPU cycles as in-progress tasks suspend themselves to wait for available memory. For many machines, if there's a way to arrange it, it would be best to balance these tasks with less memory-intensive tasks, to make full use of the CPUs.
Copyright © 2024 Einstein@Home. All rights reserved.
Already under discussion in
)
Already under discussion in this thread. Probably best to discuss there. Christian has some comments/information about memory use.
Cheers,
Gary.