No SQL .NET recap

Mongo + Norm (which gives you Link).

Why do you have all these relationships.
Make it as flat as you can.
Avoid reinventing a relational model.
Sales, get last weeks sales documents, use map reduce to get collections.

Lucene great for text searching
If you're doing an exact match on an array don't use Lucene

What if you want to search for CAT on a table.

Mongo is reliable.

Don't freak out if you replicate data (don't worry about duplicate)
Sysodb
Seven databases in Seven weeks.

Sharding - picking a key to shard on (pick a key
like first name, you'll end up with unbalanced data on nodes).

Raven (best query lang), Couch (Can write GUI app in Couch, inc mobile app), Mongo (use Normal).

Couch has terrific replication
Raven is .NET
Raven plugins are the stored procedures of NoSQL (don't quote this).

Mongo nice because you can just start without nay schema and just keep changing (refactoring) not so nice if you've very many users who expect schema.

Raven the only one that can do atomic operations across documents (the idea of a transaction). Mongo, Couch, the otters don't have notation of a transaction.

Read intensive database, NoSQL, consumer internet.

Raven (read Orin for general landscape roadmap on document databases)

Trinity from MSR (interesting to follow).

Dryad, is a distributed query store, really design for running distributed queries on really really large databases.

Consistant

Availability Partition Tolerance

See DevExpress
See Raven Screencast

Cassandra - Facebook

Cassandra extends Document store to include Column database (similar to Google's BigTable).

Memchached for user experience (Correctly designed document store is fast, use MCd for UX 1/2 second vs 1/10 second on consumer internet names a difference)

Bottom Line - choose the Right Tool for the Right Job

Session notes via @iC