Re: [LARTC] beta-release of H-FSC port for Linux 2.6

[Patrick McHardy <kaber@trash.net>]

Hi Torsten,

Griem, Hans T wrote:

>I read the abstract and introduction to original H-FSC paper.  I believe it was mostly about decoupling bw and delay.
>It seems to me HTB does this well.  When and/or why would I would want to use H-FSC versus HTB or are they direct competitors?
>  
>
Not sure what you mean with "competitors", I made this port for
fun after happily discovering that the code was very nicely
written (unlike the version mentioned in the paper). This should
not be understood as dissatisfaction with HTB, I was perfectly
happy with the entire time I used it.

So why would you want to use H-FSC .. you're right, a major
feature of H-FSC is decoupling of bandwidth and delay, but it
also offers delay _guarantees_ if configured correctly. This is
very important for streaming, VoIP, .. (and gamers of course).
I don't know if my understanding of HTB's algorithm is correct,
but it is basically a chained token bucket, so delay is directly
coupled to bandwidth. Also I believe it always tries to dequeue
"quantum" bytes at once so classes might go overlimits for a
limited period of time. This hurts delay. On the other hand,
HTB is more expressive due to priorites. So I'd say it basically
comes down to expressiveness vs. delay guarantees. However
I was able to convert my HTB config to something very similar
with H-FSC depite using priorites with HTB.

I hope that anwers your question.

Best regards,
Patrick



_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/