How to save base64 to blob of MySQL?

When should one use a traditional table-per-object schema in MySQL versus a centralized blob store with separate index tables a la Friendfeed?

• Here's Bret Taylor's post on the FriendFeed data architecture: http://bret.appspot.com/entry/how-friendfeed-uses-mysql I've heard of other websites using similar architectures.  Is this becoming the default method to scale MySQL in a rapidly growing and changing environment?  Do the benefits of ease of maintenance and flexibility outweigh the drawbacks of data duplication and loss of (easily) ensured consistency?

• Answer:

  Relational databases were originally designed to allow analysis of relationships between data using complex queries.  Originally, databases did not come under the traffic profile we experience today - you just loaded all the data into the database in batches and then your business analyst team or whoever would run complex queries to discover relationships between the data.  You didn't have millions of users connecting to your database just to retrieve their user data or whatever. In the modern internet world, databases have been somewhat conscripted into being object stores.  The majority of internet applications do not need the tight relational or integrity constraints designed into databases (not that MySQL is a shining example of such an implementation, so maybe it all balances out...) such as ensuring against data duplication or tight consistency.  If your application does not have to do with (1) the military, (2) finance, or (3) health care, you should seriously consider a datastore that doesn't use the "traditional" schema you mention in your question.  I am a pretty big fan of the data architecture designed by FriendFeed, though I haven't used it, but mostly because I think the table-per-object schema is a mismatch to modern application architectures and only in wide usage due to historical legacy reasons (i.e. tradition). (disclosure: I work at Facebook, and we recently acquired FriendFeed)