Hi Adrian,

It would have been ok if it was a connectionless transport like UDP.

However I notice a problem when I want multiple sockets to use the
same port number in particular cases. I figured that I cannot have a
server socket (accepting connections) as well as a client socket
trying to connect having the same port number, though I can allow same
port to be shared across multiple sockets in some cases by using the
socket option SO_REUSEPORT.

Thanks,
Vishwas

On Fri, Sep 11, 2009 at 9:33 AM, Adrian Farrel <[email protected]> wrote:
> Hi,
>
> For my part (I only worked as an author quite late in the process, but I was
> helping to get this through the IESG review that raised issued about TCP
> ports) I think Julien has not captured the point.
>
> The RFC is not silent about source ports because it does intend to limit the
> scope.
>
> As Julien says, it does matter which port is listened on, and this is
> deliberately a well-known port number so that it is not necessary to
> configure (or advertise) the port that must be called.
>
> But in Section 4.2.1 you will find...
>
>  Only one PCEP session can exist between a pair of PCEP peers at any
>  one time.  Only one TCP connection on the PCEP port can exist between
>  a pair of PCEP peers at any one time.
>
> One way to help ensure this is to reduce the number of available ports to
> use as the source port.
>
> Does restricting the source port cause any implementation or deployment
> problems?
>
> Thanks,
> Adrian
>
> ----- Original Message ----- From: <[email protected]>
> To: <[email protected]>; <[email protected]>;
> <[email protected]>
> Cc: <[email protected]>
> Sent: Thursday, September 10, 2009 9:57 AM
> Subject: Re: [Pce] PCE port number
>
>
> Hi Vishwas.
>
> Interesting comment. We must admit the wording is a bit ambiguous.
>
> RFC 5440 says: "The system listens to the PCEP-registered TCP port" and
> "Upon receiving a TCP connection on the PCEP-registered TCP port"; I do not
> see any behavior description for the source TCP port (good thing!).
> My guess is that there is no intend to put constraints on the TCP initiator
> port. I would interpret the sentence you mention as "using the registered
> TCP port on the PCE side, i.e. for the source TCP port for PCE to PCC
> messages and for destination TCP port for PCC to PCE messages"... A little
> cumbersome, I agree.
>
> Authors of RFC 5440, would there be an actual intend to specify otherwise?
> Implementers, any other interpretation?
>
> Thanks,
>
> Julien
>
>
> -----Original Message-----
> From: [email protected] [mailto:[email protected]] On Behalf Of
> Vishwas Manral
>
> Hi,
>
> I looked at the following in the spec:
>
> “Transport Protocol
>
>  PCEP operates over TCP using a registered TCP port (4189).  This
>  allows the requirements of reliable messaging and flow control to be
>  met without further protocol work.  All PCEP messages MUST be sent
>  using the registered TCP port for the source and destination TCP
>  port.”
>
> This has been worrying me a bit. Unlike other protocols like BGP or LDP
>
> In BGP it states clearly:
>
>  A BGP implementation MUST connect to and listen on TCP port 179 for
>  incoming connections in addition to trying to connect to peers.
>
> and
>
>  BGP's destination port SHOULD be port 179, as defined by IANA.
>
> Why do we have this restriction on source port? It is becoming a
> challenging task
> in merchant OS.
>
> Thanks,
> Vishwas
> Vishwas
> _______________________________________________
> Pce mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/pce
> _______________________________________________
> Pce mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/pce
>
>
_______________________________________________
Pce mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/pce

Reply via email to