A couple of observations. There is one in queuing theory that says that if one is putting things on the queue faster than one is taking them off, disaster is just a matter of time. It is called being "behind the curve."
We have been hearing for a long time about the ability of "cloud computing" to help one deal with the kind of peak load that may arise from event like new product announcements or open enrollment periods. Since this load was easily anticipated, one would have expected AT&T to have sufficient capacity to handle it.
The problem has been going on for three hours with few successes suggesting that we are trying to fit an extra three ours of unsatisfied demand into three hours less capacity.
However, the retries are the really aggravating factor now. The load is increased dramatically by unsatisfied retries.
The problem appears to be getting worse, not better. Early in the day, the Apple servers seem to be working. The problem appeared when one tried to "check eligibility," i.e., get data from AT&T. At this point, the transaction seems to be failing at other points.