[LARTC] Split bursty bandwidth equally

[LARTC] Split bursty bandwidth equally

[Mihai Vlad]

Hello again,

Is it possible to split a bandwidth equally among clients regardless of its
current link speed?

I have a link that can get bursty at times. At any given time the N active
sessions (the ones with non-empty queues) need to be serviced
simultaneously, each at a rate of 1/N'th of the link speed.

My case is an internet connection that oscillates between 96 kbps and 256
kbps. I cannot predict the connection speed in order to use the classical
HTB (to set up a 96/96 kbps class), because I would loose a lot of bandwidth
when the speed goes over 96kbps.

This might not be a strictly HTB related question. It doesn't matter if I
use htb, cbq pr other technique. I do not have to guarantee a certain amount
of bandwidth to one computer in the LAN, just to split the existing
bandwidth equally among the N active clients. 

I know... Someone said here "Use sfq or esfq". Unfortunately I am not very
bright and a piece of code would be excellent :) In fact I tried a lot of
setups but did not get any satisfactory result. 

Please help!


Thanks in advance,
Mihai Vlad



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

Re: [LARTC] Split bursty bandwidth equally

[Andy Furniss]

Mihai Vlad wrote:
> Hello again,
> 
> Is it possible to split a bandwidth equally among clients regardless of its
> current link speed?
> 
> I have a link that can get bursty at times. At any given time the N active
> sessions (the ones with non-empty queues) need to be serviced
> simultaneously, each at a rate of 1/N'th of the link speed.

Is this upstream or downstream traffic?

If it's upstream, what causes the variation, eg. is it RADSL and you get 
  a long queue in your modem buffer, or is it shaped later at ISP - what 
is the link type/behavior ?

> 
> My case is an internet connection that oscillates between 96 kbps and 256
> kbps. I cannot predict the connection speed in order to use the classical
> HTB (to set up a 96/96 kbps class), because I would loose a lot of bandwidth
> when the speed goes over 96kbps.
> 
> This might not be a strictly HTB related question. It doesn't matter if I
> use htb, cbq pr other technique. I do not have to guarantee a certain amount
> of bandwidth to one computer in the LAN, just to split the existing
> bandwidth equally among the N active clients. 
> 
> I know... Someone said here "Use sfq or esfq". Unfortunately I am not very
> bright and a piece of code would be excellent :) In fact I tried a lot of
> setups but did not get any satisfactory result.

Depends on the exact nature of your link, hardware and the direction of 
the variable rate. It may be impossible/possible/a bit possible more 
detail is needed.



Andy.


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