Asymetrical Routing

Peter Kline pkline at
Fri Aug 22 13:34:24 UTC 1997


Asymmetric Routing between two hosts on separate, multiply-interconnected
networks is an undeniable fact of life.  Most of the time, it works just fine.

"Problems" with Asymmetric Routing include:

- it can be hard to troubleshoot problems
- it can be hard to explain to those with minimal knowledge of routing or
internet architecture
- some applications can have problems with a path which is congested in
only one direction

but most of all:

- possibly without justification, end users think that Symmetric Routing is
the "right thing" to do, so therefore Asymmetric Routing is the "wrong
thing" to do

Tracking down problems like the one you describe can be hard; my
recommendation is that you schedule some maintenance time where you can
vary your advertisements and watch what happens from the various traceroute
sites around the net; for example, go to a Sprint-based site, and see what
the traceroute does when you stop advertising to, then when you stop
advertising to Digex.

Also, use Digex's looking glass sites ( to check your
advertisement to see if it really looks like what you think it does.

The best way for you to solve this problem is to come up with a very
specific and repeatable set of conditions under which connectivity to your
net fails.  Take that information to engineers at Digex and UUNet and I
think they'll help you out.  They can do much more with "it hurts at the
first knuckle when I wiggle my right index finger, especially since I
slammed it in the car door" than they can with "one of my fingers hurts."

good luck,

At 04:47 PM 8/21/97 -0700, Jayme Cox wrote:
>	I've been investigating a problem we seem to be having with
>asymetrical routing. We have connections from both UUnet and Digex and are
>taking full routes from both. We have had some complaints that people are
>unable to access our network and most of these cases appear to prefer the
>Digex route. However, our outgoing paths to those same networks prefer the
>UUnet route.
>	My question is: How bad is this? Is it terminaly bad or simply not
>so great? Out of order TCP packets are bad, but can simply be reassembled.
>Timout should not be a problem as both links are >= 6 Mbits.
>			--Jayme

More information about the NANOG mailing list