As stupid as it can sound, you could develop a protocol to make routers
talk each other and say how much bandwith is available in between. I
think there's no other really sane way of inbound traffic control.
Dropper techniques are a cheap trick nice for little networks. Serious
and big performance networking requires solid bases.
Think of overhead of receiving, dropping a packet, enqueing the
offending stream, waiting, listening a resend again... That looks too
much as spam :)
Regards,
Dani
irix escribio':
Hello ,
* irix <i...@ukr.net> [2009-05-27 18:12]:
But I can not understand why you are sure that traffic can only
outlet Shape
i can not understand why you want to shape outlets.
you don't understand that inbound shaping doesn't work because you
have obviously no idea how the network stack works. there is no
suitable queue inbound to do any queueing on. the ipintrq is way too
early. so to do any inbound shaping you had to insert another queueing
step, which is as clever as drinking water from the dead sea when
you're thirsty. or maybe one could rape the ipintrq somehow. but i
don't and won't rape.
by shaping the incoming traffic, I mean simple dropper without
constructing queues. All that the above specified speed dropped until
the flow becomes less than or equal to specified speed. That actually
makes CDNR, which arrears.
But it pains me to see the obvious defects in my favorite system,
interestingly, in the 6 years since I did the altq/pf merge, you're
the only one to see that "obvious defect"
and complete indifference on the part of developers to the obvious defects.
obviously the developers have no clue about what they are doing, and
the milestones they have to meet by the contract they have with you
understood the joke. Funny