[Planetlab-devel] IPv6 support for MyPLC

[Aaron Harwood]

I would like to agree with you on (1), and be optimistic that alternate
implementations of a service are just that -- but I'm not. I would  
think that
the implementations will diverge and become more or less different  
services.
This is a good thing in many ways, promoting diversity and research/ 
development,etc,
but then it can blur interoperability. Maybe I'm just scare mongering.

Unfortunately for (2) it would seem that the situation may tend to  
get worse
and never really get better.

--aaron

On 08/11/2006, at 2:53 PM, McGeer, Patrick C wrote:

> Aaron,
> There are two effects here:
> (1) Alternate implementations of existing services; and
> (2) Tunneling everything over http on port 80 because firewalls block
> other ports.
>
> For (1), it's clear that where port number is an alias for service  
> name
> (53, 110, 25, 80, etc...) using familiar ports and alternate IP
> addresses is pretty clearly the right thing to do.
> For (2), nobody is
> very enthusiastic about tunneling everything over port 80 -- but what
> choices have we?  If local admins are going to adopt the policy of
> everything-not-compulsory-is-forbidden, which they are, we pretty much
> have to do this.
>
> To take one example, I discovered today that I'm going have to tunnel
> pings over 80, because one organization that I work with blocks  
> ICMP in
> operation.  Very hard to do network measurement without pinging, but
> we're going to have to do it...(and don't ask me why people block  
> pings
> -- habit, probably)
>
> -----Original Message-----
> From: Aaron Harwood [mailto:aharwood at csse.unimelb.edu.au]
> Sent: Tuesday, November 07, 2006 7:02 PM
> To: McGeer, Patrick C
> Cc: Bound, Jim; Marc E. Fiuczynski; devel at lists.planet-lab.org
> Subject: Re: [Planetlab-devel] IPv6 support for MyPLC
>
>
> On 08/11/2006, at 12:50 PM, McGeer, Patrick C wrote:
>> As to why people want their own IP addresses, it's because they want
>> to attach services to well-known ports.  For example, there are at
>> least three collaborative DNS services I know about, and all three
>> would love to use port 53.  Similarly, lots of people want 80,
>
> Hi  all,
> Just wondering, while this is what developers want, whether it  
> would be
> desirable in the long term. Well known ports exist because they  
> supply a
> well known service; which is expected to work in a certain way.
> Perhaps a broader statement is to say that developers want their
> services to become well known and widely used; attaching to a well  
> known
> port and using a protocol that is associated with that port is a  
> design
> choice that facilitates their goal. If the service happens to be a  
> "new
> and improved" version of an already existing well known service (like
> your example above) then this facilitates integration, as well. To  
> some
> extent, and just for arguments sake, the use of IPv6 may lead to a
> situation where the IP address is the main "address" of the service;
> with port number being used to select support services for the primary
> service assigned to that IP address.
> If we promote overloading of port numbers then we might see a
> degradation of interoperability which might not be a desirable  
> outcome.
> --aaron