Stardate 20021001.2335

(Captain's log): I know that the communications problems I'm having are causing all of you trouble, and all I can do is apologize. I hope it will be cleared up relatively soon. If it doesn't get fixed, I'll be forced to look into PacBell's DSL service.

Of course, it causes me a hell of a lot more trouble. It affects you when you're trying to visit me. It affects me for everything, and I'm getting really pissed about it.

This morning I began running a small program on the server to collect some data on the quality of the link. I'm not concerned with minor things like latency; what I'm concerned with is whether traffic is getting through or not. So it sleeps 60 seconds, then attempts to ping a local server here inside the RR network ten times, and logs a time stamp and the number of pings which got answered (0-10). The result is most interesting.

This data plot shows the line quality for today, and it makes extremely clear that this is load-dependent. There's a dip at noon, and then a major dip beginning at about 4:00 this afternoon which lifts after 10:00. The vertical scale is the number of pings out of ten which get answered (irrespective of how long it took). The blue dots each represent one sequence of ten pings. The magenta line is a moving average. When the dots are at the top, on the "10" line, it means everything is getting through. When they're on the baseline, "0", it means the line is totally dead. This evening, there were several cases where for several minutes at a time I had no service at all.

It's been conclusively proved that this has nothing whatever to do with my equipment. My cable modem actually has an IP and can be pinged. The RR tech support guy did that from RR's office at my request yesterday afternoon and saw that many of them were not getting answered. That test was done totally within the part of the system belonging to RR, so it can't possibly be my fault.

RR could have gotten a tech out here sooner than Thursday afternoon. They could have sent one this afternoon but I had a doctor's appointment. They could have sent one any morning this week, but it works perfectly in the morning and there wouldn't have been anything to diagnose. Unfortunately, if the guy shows up about 3:00 on Thursday, there still won't be anything to diagnose. As a result of this data capture, I'm going to have to get hold of RR tomorrow and tell them to make sure that the tech arrives as late in the day as possible.

This also proves that it can't be a problem in the local wire coming into my apartment. Were it something like that, you'd expect the failure rate to be consistent over time. But what we're actually seeing is that it's failing during those intervals when RR has the largest amount of use (lunch, and the evening).

It's a royal pain, but there's nothing to be done about it except wait until Road Runner fixes it.