Difference between google appengine and actual big table

I know that 开发者_如何学编程app engine is implemented on big table, can anyone describe the difference between actual implementation of big table and google's implementation of big table .i.e (App engine)

---

Bigtable provides a basic key/value store, described in the paper here. Values are stored in rows and columns. Row and column keys are arbitrary byte strings. For more details see the paper. The basic operations Bigtable provides are lookups on individual row and column keys, and ranges of rows.

On top of Bigtable, there's an abstraction layer called Megastore. Megastore uses the bigtable primitives to construct a more versatile database platform. It adds indexing - using separate bigtables as indexes - and queries using those indexes. It also adds replication support. It's Megastore that provides most of what we think of as the App Engine datastore, such as composite indexes and the variety of queries the datastore provides.

Finally, App Engine implements a few things of its own on top of Megastore, such as the format of App Engine entity keys, giving each app its own datastore, and implementing certain operations like 'IN' and '!=' in an abstraction layer in each language's SDK.