Real-Time "RT" Patches Updated Against Current Linux 6.7 Development

[phoronix]

Phoronix: Real-Time "RT" Patches Updated Against Current Linux 6.7 Development

Released on Thursday were the Linux v6.7-rc2-rt1 real-time "PREEMPT_RT" patches that now re-bases the RT patches against the in-development Linux 6.7 kernel series...

https://www.phoronix.com/news/Linux-6.7-Real-Time-RT-Patches

[timofonic]

I hope some RT stuff gets merged soon, it would be good if the merge makes to be only less than 40 patches before mid 2024 at least.

[gavron]

People tend to think of RT as a specialized thing useful only to embedded systems.

RSX, RSTS, VMS, and other systems have shown that the implementation of RT primitives allows greatly improved functions in a scheduler, tasker, memory manager, heap or trash collector, priority manager, etc. Some are misled by today's implementations into thinking THESE functions are somehow all part and parcel of the same thing. Truly they are separate, and by integrating them compromises are made that makes RT not as attractive for general computing (or even RT) as it could be.

I'm hopeful eventually the integrating of RT and then an eventual breakout of the various functions along with analysis of how they compare against each other (and yes, I credit Phoronix and ML for turning this wizardry into science!) will enable different pieces all working together to make any particular workload perform at its best. Today's hardcode .config can be tomorrow's AI (ha!) concocted .config to next year's LIVE ON THE FLY modifyable kernel algorithmic choices for the above items.

E

[Volta]

LIVE ON THE FLY modifyable kernel algorithmic choices for the above items.E

Like eBPF?

[gavron]

Like eBPF?

TBH I tried to keep my language generic such as not discussing SystemD vs SySV Initscripts, etc. But since you're calling me on it..

YES, like that.

Micro-rant follows sig.

E


eBPF. YES. WHY THE HECK DOES MY CAR MAKE MORE ENGINE ADJUSTMENTS PER SECOND than the Linux kernel's backstops allow per minute? Why can't the kernel have guided parametric solutions to most task-oriented solutions? Sure, a cloud compute server might never need them once load average exceeds nproc but who cares? Sure, a desktop / laptop may never need them 98% of the time when it's just sitting there taking up space warming my "command center"." My Plex media server sure can use a nudge when it goes from 100% idle over 5 days to SNL+Simpsons+FamGuy+AmDad+Dune+whatever days. Why is this a compile-time "pick your buffet dishes now and forever enjoy them cold" thing instead of a "grab from the hot bar as you like and we'll get you fed with the right food at the right temp every time".

Yes, eBPF is one such solution. I would think after 10 years it wouldn't even be a question, and yet here we are.