Re: Reliable PPP

Fabio Poggi writes:
> I need to connect an IP host with a proper static IP address to an IP
> network through a TDM network.

Not that it matters greatly to me, but this doesn't appear to be the
right place to be asking a question like this.  The IETF PPP working
group deals with extensions to the Point-to-Point Protocol itself, not
deployment or usage.

> Moreover I need a reliable IP connection between this IP host and the IP
> network.

Please define what you mean by "reliable."

> So, I thought to set up two different PPP session (this would guarantee
> physical reliability), over different TDM links, toward the IP network and

The first problem is a matter of facilities sharing.  I'm going to
assume that by "reliable," you mean "survives a fault in at least any
one component."  In that case, do the TDM links you have in mind share
any facilities at the hardware level?  How are failures there covered?

> to manage these PPP links with an adaptation function in my IP host (in
> order to not duplicate IP traffic on IP network). This function will use,
> in Tx toward the IP network, the PPP session from which, in Rx, it has
> receive PPP traffic. When a fail occurs, the adaptation function switch on
> the other PPP session.

That can work, though I'd personally recommend either duplication or
distribution instead.  The problem with stateful mechanisms like this
-- switching back and forth between redundant links -- is that they
introduce new failure modes (e.g., having each end believe a different
link has failed), and those new modes are frequently more likely to
occur.  In other words, it sometimes makes overall reliability
*worse.*

For duplication, send the same packet over both links, and have the
peer eliminate the duplicates.  Note that PPP guarantees in-order
delivery, so this should be pretty simple.

For distribution, pick some reasonable method (such as a
flow-identifying hash) to distribute packets among the available
links, so that if a link fails, only a fraction of the connections are
affected until the failure has been detected.

> Two PPP sessions are monitored by Echo Request/Reply message.

I'd suggest looking into LQM (RFC 1989).  Echo-Request/Reply works,
but it's a bit crude and won't detect service degradation.

> The IP network will see my IP host's IP address on two different PPP link
> and it will be a task of IP network routing protocol to identify the proper
> route toward my IP host.
> In case one PPP link fails, my IP host has always an available link toward
> the IP network and the IP network routing protocol will reconverge in order
> to find out another proper route toward my IP host.

That's another way to do it.  It isn't necessary to give those links
separate IP addresses or expose them to the routing protocol, as long
as you have control over the implementations on each end.  But it is
one way to deal with the issue, and might be advantageous if there are
many different routers that have multiple links, and there are systems
behind those that can forward through any of those routers.

-- 
James Carlson, IP Systems Group                <[email protected]>
Sun Microsystems / 1 Network Drive         71.234W   Vox +1 781 442 2084
MS UBUR02-212 / Burlington MA 01803-2757   42.497N   Fax +1 781 442 1677