Mirror servers code broken.

[Ageless]

Over the past couple of days we have had multiple reports from people on Einstein@Home that they couldn't download work. At the time the problem began, one of the mirrors for Einstein was down. Yet BOINC should automatically switch to the next closest server in the list if it found the one closest to your time zone down. Apparently it isn't doing that.

Most people were able to get the work by changing their time zone, so BOINC was forced to use the mirror closest to the new time zone.
Others were able to download the work by hand, by adding the full task name and link to one of the mirrors in a browser.

It's platform independent as well, with Macintosh, Linux and Windows machines affected. BOINC version numbers all 5.10s, mostly 5.10.28 and 5.10.30

So is it possible that the code has been broken for a while, but no one noticed because the mirrors were all up and running? Or is there something else we'd have to look at?

[Nicolas]

The download feature could be greatly improved. For example, pick a server (closest or random or whatever). If it fails to connect, pick another. If it succeeds, but the md5 or digital signature don't match, pick another, and make sure the server that gave a corrupted file isn't ever tried again for that file.

Is the timezone currently matched only based on -/+ hours from GMT? That doesn't quite help if server is in the north and client is in the south or similar. Being in Argentina, I could download faster from a server in Brazil (one or two hours off me) than from one in USA with my same timezone. Can't the client know the user's country based on IP address, or some setting on the machine? Timezone looks like a very inaccurate way to locate a user.

For big files, it could even download from two random servers of the list at the same time...

[Nicolas]

And to provide info like server location and preference, how about using a ​standard XML format...

[davea]

Probably fixed in 6.6.11