Biblio.com down for over 24 hours and running...the revenge of unintended consequences...

My favorite online aggregator, biblio.com, had their host blow up (figuratively). Apparently, they had a power outage and the large UPS that was intended to keep the servers up and happy did not work (I'm certain there will be joyful discussions about this). The problem with large, dynamically served database driven webservers is that they do not like having the proverbial plug pulled...it makes them cranky and, on occasion, uncooperative as to being revived. The host, as I understand it, has power again and one or two other book sites hosted there have returned...unfortunately Biblio.com remains down. Presumably this is due to having to rebuild the data base, but it raises at least two important issues related to running large...er...important servers:

Backups: Realtime remote backups that can be turned to when the unexpected happen are good things. Sort of like insurance with less annoyance.

Peering: Having multiple backbones coordinated so that when one line gets too slow the server is automatically switched to the fastest line (load balancing). This is tricky as many hosts claim to have "multiple connections" to the Internet while *very few* provide true "peering". Speed is good, stability through a node loss is better.

Mirroring: Having duplicative servers at different locations (backups for grown-ups). The mirrors serve as a back-up, AND support stability and connection speed.

On the other hand, you can run everything off one server in and roll the dice. While Biblio.com is down, I suggest reading Edward Tenner's book, "Why Things Bite Back: Technology and the Revenge of Unintended Consequences."