[LARTC] guarantee package delivery

[LARTC] guarantee package delivery

[Vladimir S. Petukhov]

Hi to all!

Sorry for my English :)

The problem:

We have a shaper software based on tc linux shaper/filter. This software 
('shaper') work with one interface ('eth0') with fixed bandwidth and limit 
user (above 100 user each time) bandwidth accordingly user's tariff plan. 
User traffic filter based on destination (user) ip. There a lot of connection 
types: OpenVPN, ipip, PPP...
  But we want to provide 	guaranted bandwidth to ONE userspace application 
(filter may be applied by port number). Moreover - packets must not be 
dropped! One of the obviously decisions: Module (kernel) must inform 
userspace about current bandwidth or data amout, that programm can be send 
this moment.
  How it can be done?

Thanks....
_______________________________________________
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

Re: [LARTC] guarantee package delivery

[Andreas Klauer]

On Sun, Jan 15, 2006 at 12:41:31AM +0300, Vladimir S. Petukhov wrote:
> Moreover - packets must not be dropped!

Sorry for this useless answer, but... How strong is this condition?

I mean, even if you don't drop a packet locally, it can still be 
dropped by the target machine, or by one of the routers in between. 
You have no influence on that whatsoever, so no matter what you do, 
your application must be able to handle dropped packets.

If you think about it that way, is it still critical when a packet 
gets dropped locally? If not, you could just do this the usual way.

> One of the obviously decisions: Module (kernel) must inform 
> userspace about current bandwidth or data amout, that programm can be send 
> this moment.

Does the kernel even know about that? 

Regards,
Andreas Klauer
_______________________________________________
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

Re: [LARTC] guarantee package delivery

[Vladimir S. Petukhov]

>> Moreover - packets must not be dropped!

>Sorry for this useless answer, but... How strong is this condition?

> I mean, even if you don't drop a packet locally, it can still be 
> dropped by the target machine, or by one of the routers in between. 
> You have no influence on that whatsoever, so no matter what you do, 
> your application must be able to handle dropped packets.

No. There are no routers and no any other hosts between server and client 
machine. There are only ... air and space :) . I talk about sattelite link 
with 100% qality within fixed bandwidth. In way from server to client only a 
shaper can drop packets.
 This programm (client-server) use this characteristic (this and some other) 
and "accellerate Internet access". But any packet loss entail speed fall and 
a lot off "land" high-proced traffic. 

> Does the kernel even know about that?
 Of course. We use HTB to separate speed between user, and may be logic 
implemented, that guaranted package delivering to the network adapter. That 
all we need: do not filter THIS traffic at all - talk about availible 
"traffic" to userspace programm (using tokens, e.g. or in some other way).
_______________________________________________
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc