Assuming you are using the term "degraded" to mean the link has a high bit error rate but is otherwise up (so the link is not flapping, it's just unuseable) you have a real challenge. The problem is that the short hello packets used by EIGRP to test the link are getting through frequently enough to keep the link up, while full size packets have a very high failure rate.
Side note: Consider a link with a 10e-4 BER. The probability of a 50 byte frame getting delivered is 95% while the probability of a 1500 byte frame surviving the journey approaches zero.
If this is your problem, welcome to the club. Short term, the only solution is to shut down the link until the phone company fixes it. Longer term, the "correct" solution is to use PPP Link Quality Monitoring to detect a failing link and shut it down at the link level.
Unfortunately, while Cisco supports LQM, when I last tested it, their implementation made it worthless. The link gets shut down when the percentage of failed frames exceeds a threshold, the problem is that the percentage is calculated on the total number of frames since the link was initialized rather than a running average. As a result, a link which fails after a year of heavy use will take a long time to reach even a low threshold.
Another possible solution (this problem has been around for a LONG time) is to use a routing protocol, such as IS-IS, which supports padding the hello packets to an arbitrary size. This approach is another good idea which turns out to sound better than it works. As you can imagine, the impact on efficiency makes this approach questionable on slow links, where BER problems are most likely to occur.
It also turns out to be too slow to react, as the link will stay up as long as at least one out of every three hellos gets through, while the performance impact typically becomes unacceptable when the failure rate is still in single digit percentages.
Currently, the only viable tool is monitoring error rates using a network management tool and getting the telco to repair the links before the BER gets high enough to be significant. Easier said than done. Meanwhile, you can try PPP LQM (if PPP is supported on your interfaces) and beat up on your Cisco account team to fix the implementation so it is useful (which may have already happened, although I have not noticed any announcements to that effect, which is why I suggest testing first).
Good luck and have fun! Been there, done that, been burnt :-[
Vincent C Jones
www.networkingunlimited.com
