I apologize for not having participated lately to the work of the WG, being
taken by others important (to my familly) tasks. I am trying to bring me up
to speed now the fight seems over :-).
An important point came accross recently at TFIPv6 and through some words
of Tom Leighton (Akamai). It could be summarized this way "are routing and
addressing" to be related? Today the routing is fixed tables dependant and
addresses are ISP dependant. There are aound 140.000 routes in these
tables. Rumors (? may be someone knows better) have that over 200.000
routes there might be instability.
I observe that many problems currently discussed over IPv6 would disapear
would the SLA (Second Level Address) be totally independant from the TLA
(Top Level Adress). This would mean that the begining of the IPv6 address
would be used identify the physical network (as the TLD indentifies the
virtual network). And the SLA would identify the physical System (as does a
telephone number which uniquely identify a mobile whatever the local
network used to call it - for example abroad).
This would for example permit to support Catenet (the archtecture of Louis
Pouzin they considered for a while where the inputs of all the links are
concatenated on a per port basis, so you may receive data through a non
DoSed ISP). This would simplify the knowledge and the organization of
dynamic tables for routing.
To test the concept an archtecture is necessary without redesigning
routers. Could the architecture be a network or OPES servers used to route
datagrams over OCP.
1. let suppose that I use 60 or 100 OPES servers.
2. Each of them has physical links to Hosts and to other OPES.
3. each time a datagram is sent a dispatcher looks at the destination IP
address and if it is among the connected Hosts it builds an OCP path to
reach that host as a list of {OPES server:port nr}.
a) I understand that OCP could support it
b) I understand that OCP could support many additional services such as
killing a blocked connexion?
c) this is software, but I would be interested in comments about the
probable speed?
d) how OCP could support an exploration of the OPES servers network at
start (or on demand) to kown the most simply the topology of the network
and of the Hosts ports (this can be a simple application, but would you see
some OCP features that would help verification, speed, etc.?)
e) obviously this could be partly hard coded and gain speed and stability.
CommentsN
One of the interesting feature I suppose OCP would support easily would be
a fast and no loop routing in having the current machine removed from the
list of the {OPES:Ports] when the datagram is sent. This simple service by
the current OPES server would create a true pipe, OCP being able to report
the end success and the on-path failures and request alternate routing. The
OPES servers could bufferise the data until they are told the next node
forwarded them (or the first server until data reached the end). If the
next node failedn an alternative path could be requested and no datagram
could be lost. If I am right OCP would also make sure there is no intruding
datagrams?
jfc