Re: Not particularly interesting
"It sounds like a layer over big table..."
Only on the surface. BigTable is a log-structured merge-tree. It is very, very highly optimised for random IO, to the point where doing any kind of scan at all is going to slow performance to a shocking crawl. Spanner sits in the middle ground between immutable columnar data and pure key-value stores. Not as fast for raw analytics, and not as fast for raw IO, but mutable and analytical at the same time.