Man these guys are smart. http://www.slideshare.net/kevinweil/nosql-at-twitter-nosql-eu-2010
Talks about Twitters many different approaches to storage (Hadoop, Cassandra, FlockDB) and the tools used in querying of that data to answer questions (Pig)
Watched this presentation about how Sourceforge chose MongoDB for their customer facing webapp. You know, the one you go to download Azuerus and all those open source apps from
Sourceforge chose Mongo because it offered them high read performance although write was sucky but not needed for their app. They learnt that even though you can put things in one document, you didnt need to retrieve everything all at once and how they easily used up all network bandwidth between the web and mongo server. Additionally they found they didnt need memcache in between their persistence and the webservers as Mongo was fast enough to serve out the data as it was.