[LARTC] University Project: QoS with Ai

[LARTC] University Project: QoS with Ai

[GoMi]

Hi there guys, i am on my last year of career, and as my final Project i am
interested in doing something about QoS.

The thing is, I have a couple of QoS solutions working, and since the
beginning I though it was great but it lacked some kind of dynamicity. 

Let me explain myself.. I was thinking in creating a classful queue that
based on some parameters (kind of users, bw, packets, etc...) could "learn"
in some way the kind of traffic passing through the box, and change the
parameters of the classes, classifying also not only the packets, but the
users.

For example there could be p2p users, HTTP users, etc.. and if the queue
itself could create classes for those users, maybe increase productivity.


What do you think? Am I talking bullocks or it makes some sense? All critics
welcome :)


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

Re: [LARTC] University Project: QoS with Ai

[Damjan]

> The thing is, I have a couple of QoS solutions working, and since the
> beginning I though it was great but it lacked some kind of dynamicity. 
> 
> Let me explain myself.. I was thinking in creating a classful queue that
> based on some parameters (kind of users, bw, packets, etc...) could "learn"
> in some way the kind of traffic passing through the box, and change the
> parameters of the classes, classifying also not only the packets, but the
> users.
> 
> For example there could be p2p users, HTTP users, etc.. and if the queue
> itself could create classes for those users, maybe increase productivity.
> 
> 
> What do you think? Am I talking bullocks or it makes some sense? All critics
> welcome :)

Very good idea, 
I don't know how hard it would be to implement it, though.

-- 
damjan | дамјан
This is my jabber ID --> damjan@bagra.net.mk <-- not my mail address!!!
_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] University Project: QoS with Ai

[Ethy H. Brito]

On Tue, 24 Aug 2004 21:18:03 +0200
Damjan <gdamjan@mail.net.mk> wrote:

> > For example there could be p2p users, HTTP users, etc.. and if the queue
> > itself could create classes for those users, maybe increase productivity.
> > 
> > 
> > What do you think? Am I talking bullocks or it makes some sense? All critics
> > welcome :)
> 
> Very good idea, 
> I don't know how hard it would be to implement it, though.

This idea was once my Master Science work proposal. I was crucified by the Phd's saying that AI does not goes well with packet classifying at high speeds (something to do with 'too big decision tree'). You can read 'high speeds' like 100+ Mbits networks.

My 2 cents.

All the best

-- 

Ethy H. Brito         /"\
InterNexo Ltda.       \ /  CAMPANHA DA FITA ASCII - CONTRA MAIL HTML
+55 (12) 3941-6860     X   ASCII RIBBON CAMPAIGN - AGAINST HTML MAIL
S.J.Campos - Brasil   / \ 
_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] University Project: QoS with Ai

[Brian Carrig]

GoMi,

For an ongoing project on QoS we looked at something similar but were weary of 
complex decisions being made at high speeds, particularly as most users questioned 
would prefer to have a guaranteed equal share of the bandwidth. Currently we are 
working on a GUI that allows users to mark their traffic as belonging to Gold/Silver or 
Bronze classes of service. They are then charged appropriately per volume. The 
results obtained by this approach thus far are quite satisfactory.

Regards
Brian

On 24 Aug 2004 at 18:47, GoMi wrote:

> Hi there guys, i am on my last year of career, and as my final Project i am
> interested in doing something about QoS.
> 
> The thing is, I have a couple of QoS solutions working, and since the
> beginning I though it was great but it lacked some kind of dynamicity. 
> 
> Let me explain myself.. I was thinking in creating a classful queue that
> based on some parameters (kind of users, bw, packets, etc...) could "learn"
> in some way the kind of traffic passing through the box, and change the
> parameters of the classes, classifying also not only the packets, but the
> users.
> 
> For example there could be p2p users, HTTP users, etc.. and if the queue
> itself could create classes for those users, maybe increase productivity.
> 
> 
> What do you think? Am I talking bullocks or it makes some sense? All critics
> welcome :)
> 
> 
> _______________________________________________
> LARTC mailing list / LARTC@mailman.ds9a.nl
> http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/


-- 
Brian Carrig
Department of Computing & Networking
Institute of Technology, Carlow
Tel. No.: +353 59 9176209
_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] University Project: QoS with Ai

[Marcin Sura]

Witam

There is similar, finished project that also uses concept of neural networks.
http://rainbow.mimuw.edu.pl/SR/prace-mgr/kaszynski/

Unfortunately, this is all in polish.


-- 
Pozdrawiam
 Marcin                            mailto:slacklist@op.pl

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

Re: [LARTC] University Project: QoS with Ai

[adi]

or use this link :
http://www.tranexp.com:2000/InterTran?type=url&url=http%3A%2F%2Frainbow.mimuw.edu.pl%2FSR%2Fprace-mgr%2Fkaszynski%2F&text=&from=pol&to=eng

adrian

> Witam
>
> There is similar, finished project that also uses concept of neural
> networks.
> http://rainbow.mimuw.edu.pl/SR/prace-mgr/kaszynski/
>
> Unfortunately, this is all in polish.
>
>
> --
> Pozdrawiam
>  Marcin                            mailto:slacklist@op.pl
>
> _______________________________________________
> LARTC mailing list / LARTC@mailman.ds9a.nl
> http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
>

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

Re: [LARTC] University Project: QoS with Ai

[Damjan]

> > > For example there could be p2p users, HTTP users, etc.. and if the queue
> > > itself could create classes for those users, maybe increase productivity.
> > > What do you think? Am I talking bullocks or it makes some sense? All critics
> > > welcome :)
> > 
> > Very good idea, 
> > I don't know how hard it would be to implement it, though.
> 
> This idea was once my Master Science work proposal. I was crucified by the Phd's saying that AI does not goes well with packet classifying at high speeds (something to do with 'too big decision tree'). You can read 'high speeds' like 100+ Mbits networks.

You don't need to make an AI decission for every packet transmited,
you just need the AI to sligthly reconfigure your TC setup, based on
stats already gathered by the TC system.

-- 
damjan | дамјан
This is my jabber ID --> damjan@bagra.net.mk <-- not my mail address!!!
_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/