Announcement

Collapse
No announcement yet.

Oracle Developing "bpftune" For BPF-Based, Automatic Tuning Of Linux Systems

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Oracle Developing "bpftune" For BPF-Based, Automatic Tuning Of Linux Systems

    Phoronix: Oracle Developing "bpftune" For BPF-Based, Automatic Tuning Of Linux Systems

    Well, here is something nifty being worked on by Oracle. Oracle engineers have been developing "bpftune" as a new always-on, automatic tuning of Linux systems -- in particular, the many different Linux kernel tunables available and this tuning system leverages the kernel's Berkeley Packet Filter (BPF) observability features to carry out its work...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    Michael Testing...?

    Comment


    • #3
      As opposed to tuned?

      Comment


      • #4
        So how to you "tune" f.i. the neighbour table size without any additional information? Looking at the code, what it does is simply increase the GC threshold without any bounds. You could of course achieve the same by simply removing all limits from the kernel, but there's a reason why they exist, because it makes for an easy DoS otherwise.

        As this thing is according to their website supposed to be zero-config, I really don't see how it can make decisions any better than what the kernel can do itself.

        Comment


        • #5
          Very exciting! Sounds like a very cool program. Something like this paired with a custom kernel tunned to your hardware would be amazing!

          Comment


          • #6
            this is a cool project, reminds me of another oracle project back in the day (well sun project), which was se toolkit's virtual adrian. the basic premise of obvervation based tuning is a great idea, but I'd hope there's one performance governor for the kernel that can oversee all this activity, you got all these agents like irqtune, p-state/c-state governors, and now this very often they should be under control of the system operator (user or job scheduler) to provide alignment goals. like for example, when I'm playing a game, power limits shouldn't matter at all and network buffers should be prioritized and limited, but when i'm just surfing phoronix, you can slow things down for efficiency. I've heard that google does datacenter level power management that tweaks how systems behave due to grid level considerations.

            Comment

            Working...
            X