[Gllug] sorts of ports

[Bruce at mailer.cafod]

>>>>>>>>>>>>>>>>>> Original Message <<<<<<<<<<<<<<<<<<

On 9/4/01, 9:50:32 AM, "Paul Brazier" <pbrazier at cosmos-uk.co.uk> wrote 
regarding RE: [Gllug] sorts of ports:


> I've always been a little puzzled by ports as well.
> It seems like an unnecessary abstraction layer between the IP packets
> and the "service" required.
> I suppose there must be historical reasons for it (perhaps there are
> logical ones also?) but why couldn't the IP packet header just request
> http/ssh/ftp or whatever directly (e.g. by including the string "http")
> instead of going via an arbitrary "port" number? Is it just to save
> space in the IP headers?

The ip protocol knows nothing about web services.  It musn't know about 
web services.  Web services are run at a higher level.  The ip protocol 
simply offers a way for two hosts to communicate.  It knows nothing about 
the applications that use it to communicate.  This is why ports are 
necessary - each sending or listening application takes a port.  Then the 
ip stack knows which application should receive incoming packets.

That's why it's called a stack - each layer deals with a specific set of 
tasks and level of complexity (although the ip protocol suite has a 
simpler and messier stack than the OSI reference model).

The system of known ports, where standard services use specific ports, is 
a hack from the early days of the net.  Many alternatives have been 
suggested, often involving extensions to DNS (your cue, Alex).  Inertia 
has kept the status quo.  Any solution will have to be robust and 
backwards-compatible.

Building firewalls could be much easier under such a solution, for one 
thing.

-- 

Bruce


-- 
Gllug mailing list  -  Gllug at linux.co.uk
http://list.ftech.net/mailman/listinfo/gllug