|
From: Jonathan D. <jd...@wu...> - 2005-03-03 05:05:31
|
Another serious issue we face is that of replication. This really has two facets: Backups: what if a server has a hardware failure, or, for one reason or another, disappears forever? It would be best if the user data is (also) stored somewhere else. Temporary failure: what if a server goes down temporarily? Ideally, the system should be able to handle this situation as well. I don't have many good answers for these. Some possible answers (which does not necessarily cover both facets): - Each server sends us a backup of itself every X time period. - Each server replicates itself (somehow) to another server. In case the first server fails, the second server starts serving the users of the first server, in addition to its own. If the first server comes back, the second server stops serving those users. If the server never comes back, the second server moves some users somewhere else. (This is basically some kind of dynamic cluster-to-cluster user-handling system, where each system pushes users around as necessary. Also, the data is always in at least two places.) Lots of fun stuff to think about! ------------------- There was a significant typo in my last e-mail: This perfect for "clustering" where each server is responsible for any number of servers. => This is perfect for "clustering" where each server is responsible for any number of users. ------------------- --JD |