> BGP is a better idea (of course).
> You might also consider using BGP.
> And have I mentioned BGP?  8-) 8-).

Whether to use BGP/OSPF is orthogonal to multipath use.  Both
OSPF and BGP allow you to install multiple next hops.  Adding
multipath support requires, at a minimum, changing struct
rtentry to store multiple `gateways' (which are really next
hops).  You also need to fix up code to do the right thing
when adding a new route or deleting an existing route (for
example, a route is not considered dead until paths all next
hops are dead).  For forwarding a packet you can always use
the first next hop to start with but very likely you'd want
to change the forwarding code and use some policy to pick a
next hop.  Typically you'd want to use the same next hop for
a given source so as to not mess up TCP RTT calculations.

<start meta discussion>
Seems to me that the needs of client, server and router
machines are different enough that the networking stack code
needs to be much more flexible.  One can always put in the
most elaborate solution and just not use fancier features for
client machines but then every one pays the cost of
complexity.  Not sure what is the right thing to do here.
Anyone care to pontificate?

The same situation exists w.r.t. a number of other features.
</start meta discussion>

-- bakul

To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-current" in the body of the message

Reply via email to