You guys might find this interesting. First is my latest email to NOAA and the second is their very prompt response:
My email:
You know, I hate to even mention this but if you look at my data for the last hour or so you will see that my data wasn't being recorded for a period of time. I mention it because it has happened before, and it isn't just me, and it isn't just Rainwise. Some other guys on wxforum.net have seen it too. The grassy knoll types think it may have something to do with some Internet servers in Texas. I don't know what it is, but it happens to me a few times a month. The data inexplicably stops being reported to CWOP, and then just as inexplicably it starts up again, usually within a couple of hours. I don't really expect an answer, but I thought this was an opportunity to put this problem in front of some intelligent eyes. Thanks again for all your help.
NOAA reply:
Hrmph. The internet servers in Texas moved, with me, to Oklahoma about 3 years ago. I have had more network interruptions here, but I'm hoping a new set of virtual machines I'm migrating to will help fix that. Different cloud different building, different network. When they were in Texas, I controlled the infrastructure at Texas A&M a whole lot more than I can here.
And, the overall number of stations doesn't drop nearly as much as a lot of folks want to think. I will keep a watch on it, though.
Our biggest, documented problem is, some servers do go down and take time to ramp back up. cwop.aprs.net and rotate.aprs.net are two "rotation" addresses. The order and server that come up are different each time. And the problem is Windows, overall. While Win 7/8 are better, they still don't do internet properly. If an application is running, every time it has an update to send, it SHOULD ask for the address using a system called the Domain Name Service, and will get a number of options. It should store each IP address in an array, and try them in order (the order will generally change each time it's queried, so while it's possible you'll get the same IP address for a server twice in a row, statistically that change is slim. The application SHOULD then try each server, in sequence, to send the data update, until it finds and logs into one.
Some Windows apps get the first address in the list, lock on to it and use that one 'til the application is restarted. Some will get the first address, log in, and never let go. We've taken steps to disconnect 'em and these generally will work better than the first set. Few actually record all the addresses, and step through 'til they get a hit, send the data, then disconnect. Which is what the spec says they SHOULD do.
Thanks for bringing that up; I'll watch a bit more closely, and hustle on the changeover for my personal servers.