On Tue, 3 Mar 2015, Juliusz Chroboczek wrote:
I still think there needs to be quite a lot of work done on APIs and best
common practices in order for applications to do the right thing so this
kind of renumbering event works. Most likely it's going to require a FOSS
library that will act as a middle layer between the application and the
network
What are the applications that you think would benefit from that?
UDP-based applications, mind you, since MP-TCP works marvelously for TCP.
EVERYTHING that is not using TCP. Which is a lot. I don't want sessions
that last more than a few seconds to rely on the address, anywhere.
I think getting thoroughly acquainted with previous art is necessary.
I'm sure there are other UDP-based applications than just Mosh and
µTP-based BitTorrent that can deal with changing addresses, and we don't
want to build something that's either too general, not general enough,
or even both at the same time.
That's why I said 5-10 man-year effort.
I don't know on what level to solve this best. Since it requires some kind
of authentication, perhaps it should be done by in a similar fashion to
IPSEC but be done on a per-session basis, not per-IP.
Also, TCP is hindered by often being included in the operating system and
not under the application developer control at all. This is fine for most
applications, but the larger ones with special needs might want to do
something differently. Looking at for example QUIC, they went down the UDP
route to fix this problem.
So what I envision is a standardised protocol that could be implemented as
a library on the host, be cross-plattform, probably run over UDP (at least
short term), and combine some of the functionality of IPSEC and SHIM6 to
enable authentication, encryption and address independence.
--
Mikael Abrahamsson email: [email protected]
_______________________________________________
homenet mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/homenet
