"Bob Puff@NLE" wrote:
>
> > I can only suspect you are experiencing the rule
> > which states: "a packet from netX cannot be routed through netY and back to
> > netX" [EXCEPT if the final destination is netZ]*
>
> One more thought. If a http request is made from netX to a machine on netX,
> and the machine generates reply packets sent thru netY to netX, aren't we
> talking about two entirely different packets? Or are reply packets somehow
> tied to their request packet, sort of like what IP MASQing does?
>
> Bob
Each direction is distinct and separate; though they generally follow the same
path.
Let's look at some examples...
Topology:
r2--------r4----s2
| |
| |
s1----r1--------r3
1) All in same net: any path combination of [s1>r1>r2>r4>s2 | s1>r1>r3>r4>s2]
<--> [s2>r4>r2>r1>s1 | s2>r4>r3>r1>s1] is valid; "ping -R" will often show the
different paths if r1-r2-r4 has the same routing metric as r1-r3-r4 (unless the
"flow" is nailed to one path; even in this last case, the "to" and "from" paths
could be different depending on the router implementation. Careful examination
of the "ping -R" outputs is required.
2) All but r3 in same "[ABC] net"** (your reply example): in classful routing,
traffic would NEVER go through r3. In classless/classful combo (ISPs can
implement this combo by accident), it depends... All classless, r3 could be
used.
3) s1 in netA, r[124] in netB, r3 in netC, s2 in netD (all classful): r3 could
handle traffic between s1<-->s2 because this is covered by the "exception"...
depends on metrics at r[14].
** Using [ABC] *very* loosely... really mean classful, though not necessarily
restricted to /8, /16 or /24 masking...
Clear as mud...? :^)
Pierre
