In part, I was thinking that this sort of thing could make experimenting with
codel or HTB tweaks much easier. If the classification logic could be written to
be compiled into eBPF filters, then changing the classification would be just
loading a new module. And since it's compiled and loaded into the kernel, I'm
hoping that the performance would be on the same order as C code (even though
it's interpreted by the eBPF engine), so it wouldn't have the large performance
hit that we have with the current throttling code for example.
David Lang
On Thu, 11 Dec 2014, Dave Taht wrote:
oh the bpf stuff and jit work is extremely interesting. There is a
llvm compiler for it too, now.
nftables is also interesting but hard to wrap my head around and
expressing things in it seems only slightly less hard that just
adopting bpf directly.
On Thu, Dec 11, 2014 at 4:28 PM, David Lang <[email protected]> wrote:
http://lwn.net/SubscriberLink/625224/2edd6566c16cf76d/
A new capability is being added to the 3.19 kernel that allows writing BPF
programs that run in the kernel that can do things like classifying traffic.
Right now all this can do is to store information to be retrieved by
userspace, but I wonder if it would be useful to use something like this for
traffic classification for queue selection, traffic throttling, etc?
_______________________________________________
Cerowrt-devel mailing list
[email protected]
https://lists.bufferbloat.net/listinfo/cerowrt-devel
