Path: utzoo!attcan!uunet!snorkelwacker!usc!sdd.hp.com!decwrl!ucbvax!SH.CS.NET!gwilliam
From: gwilliam@SH.CS.NET (George Williams)
Newsgroups: comp.protocols.tcp-ip
Subject: Re: tcp source port number assignment in new tcp connection
Message-ID: <9009060814.AA16012@ucbvax.Berkeley.EDU>
Date: 5 Sep 90 17:48:26 GMT
References: <15096@shlump.nac.dec.com>
Sender: usenet@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 32


Hello,

I don't know what TCP/IP interface you are using but unless you are writing 
your own TCP implementation source port assignment should be a 'local',
i.e. OS or TCP interface running on you machine, issue.

All interfaces I've seen on DEC machines , and others, have returned this
identifier; usually with the option for you to specify one....and even that may
not result in what the remote TCP sees as your source identifier. On the
interface level it is just a logical assignment and does not necessarily map
to a 'real' TCP port. Maybe QIO stuff is different but even that returned a 
logical channel number if i recall correctly. If security is the issue then
that's another discussion.

In any event:

 () If you have a choice in the matter let your interface to TCP assign this.
    Less book keeping and cleanup.

 () If you are writing an interface and or your own TCP then take a look
    a what Excelan, Wollangong, DEC, and others have done already in this
    area....

 () There were some discussion in this newsgroup ( whos right/wrong ) regarding
    two vendor's port assigments above 1000 which referenced an RFC, about a 
    month ago. Impression I got was that the RFC came out the winner...no hard
    fast rule just agreed on and established convention in this area.


 Good luck,
 George Williams