On 10/08/13 03:20, Alex Elsayed wrote:
Regarding ETSN, I'd recommend looking up RFC 1078. It does much the same thing, already exists... And is not particularly beloved. A good bit of what is seen as wrong with it is that, like ETSN, all connections use one port. This does make it convenient to allow through. It also makes being selective about what is allowed through *impossible* and is thus a huge security issue.
Yet I've seen users asking for exactly that (using one port) a lot of times. Also, have a look at HTTP. It's a single port and it's used for virtually anything.
Anyway, as for TCPMUX, it's exactly the same as ETSN and using port 1 would be cute :)
However, there's one concern to address:The only TCPMUX implementation I've found is in inetd and it responds to a new connection by launching a new service process. This is how the things used to work in 1988, however, today we just want to pass the new connection to an existing process via SCM_RIGHTS or similar.
Thus, a different daemon is needed which would clash with inetd (if the latter is run in TCPMUX mode).
So, the question is: Is anyone using TCPMUX today are can we consider it an unused protocol ready to be re-used for our purposes?
As an alternative, I was thinking about using WebSockets for similar purpose (use URLs as service names), but it faces a similar problem: running on port 80 any standalone service would clash with web server that may be running on the same box. The other option would be to add particular code to the web server itself, but doing it for *any* web server doesn't seem realistic.
Thoughts? Martin
