mitchellharper12 / folding-scripts

Helper scripts used to optimize Folding@Home and Rosetta@Home cores

An assortment of scripts used to wrangle Folding@Home processes. See my blog post Making Every Cycle Count in the Fight Against COVID for a deeper introduction and background on this repository. These scripts are provided as reference only, and commented out lines are included in scripts to leave a history of techniques attempted.

Important files:

dumber_pinning.sh

This script is currently runs as a cron job on my dedicated folding box and handles assignment of folding threads to CPU cores. It additionally sets compute and IO scheduler hints on managed processes. It is slightly less complicated than the script I originally published called better_pinning.sh.

better_pinning.sh

A script (currently very tailored to my TR 1920x machine) for pinning folding at home cores to specific threads, and moving all other threads appropriately. This was my first attempt at writing a management script, and it has been obsoleted by dumber_pinning.sh.

irq_move.sh

A script for moving kernel threads and IRQs around to processors you specify (mostly so they don't interrupt folding threads).

FahCore_22

A bash script interposed between the F@H manager and the GPU folding core that forces all memory allocations to be bound to NUMA node 1 (see blog post in introduction).

FahCore_a7

A bash script interposed between the F@H manager and the CPU folding core that forces all memory allocations to be bound to the node upon which a thread is running. Also sets up the LD_PRELOAD for not calling sched_yield.

sched_yield_hook.c

A C file that, when built using the provided Makefile, produces a shared object that can be preloaded that always causes the libc wrapper over the sched_yield syscall to return 0 before every yielding to the kernel. Also optionally can be built in a configruation that allows for selective yielding, but there's no evidence there's any use case where that leads to a performance gain.

Makefile

Builds the shared library described above using GCC settings tuned for the Zen v1 microarcheticture.

License:

Dual licensed under the MIT license and the AntiLicense, see LICENSE.md.