Well, I do not want to kick against any religion, but in grid computing (like for example Java Spaces) memory is the new harddisk. Since memory gets cheaper and processors are becoming faster I think there IS some cause for a memcached like mechanism to be used in a distributed, replicated way.
I am thinking of a configuration where different memcached servers maintain a copy of themselves on secondary (or even more) machines. In the unlikely event of one primary server going down the client just refers to a secondary server. Memcached entries might replicate themselves after a put or replace operation is completed. Particularly when Memcached is used for sessions it is unlikely that a get operation will follow in the instant that a put or replace operation has completed on the same entry. I realize that this takes up extra network bandwidth and cycles but it definately offers an interesting mechanism for load balanced webservers hitting the same cache to retrieve session data. And if I am not mistaken such an architecture would still be of O(1), that is it still would grow lineairly with web traffic. Anyone interested in a discussion about this? seems a nice optional feature for Memcached.
