[Bug 289333] [Feature Request] HFSC overhead calculation

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=289333

--- Comment #6 from Daniel Engel <[email protected]> ---
(In reply to Oleksandr Kryvulia from comment #5)
I don't see a whole lot ofmuch potential in rate-limiting the number of
packets.  The mix of packet sizes never skews consistently in one direction or
the other.  It varies by time of day and current user pattern(s).  I will
confess to not knowing much about ng_car and netgraph, but it appears to be
mainly an API.  It's not clear to me how writing a custom solution in user
space is any significant improvement over writing a custom solution directly
into the kernel.  

So far, all I see are one-dimensional solutions to what is (at least) a 1.5
dimensional problem.  The fact that Linux has a direct analog for the feature I
am asking for implies to me that it can't be readily synthesized from some set
of lower-level features.  

Based on my reading, it seems that HFSC already has almost all of the
infrastructure to support a two-dimensional solution.  Namely, the x and y
coordinates for the service curve(s).  It seems that every time that
altq_hfsc.c calls m_pktlen(m) -- and there are only a small handful of such
calls -- that it would be relatively easy to add a configurable parameter
and/or helper function that simulates downstream overhead.  

A couple of additional parameters could even allow HFSC to simulate the effect
of the 48-byte ATM boundaries and encoding, although the conventional 15%
derating in line throughput probably remains an adequate approximation.

-- 
You are receiving this mail because:
You are the assignee for the bug.