Autocompletion of strings from a very large database. Search by string?

Question

SQL

An interesting article about FIAS appeared on Habré the other day .
On fias.nalog.ru there is an example of a form with a search for all addresses.
The simplest example of implementing a street-only search is obvious: "SELECT * FROM street WHERE name LIKE '?%'" (with an index on this very name, which will only be used if we search for the pattern '?%').
And what are the more efficient and productive solutions? Lucene? Sphinx?
If it doesn't make it difficult, poke your finger, what to read.

Answer 1

I did this on Sphinx, yes, very quickly, and I advise him.

Answer 2

Indeed, an engine tuned for such tasks will work much faster here. I only came across Sphinx, so if it is possible to install software on the server (as well as start services), then I think it's worth looking in his direction. Installing and configuring is a matter of one day. And in the future, you can strain it for some other services of the site.

Answer 3

And what is unproductive in LIKE something%? The index is used.

Answer 4

The first thing that comes to mind is indexes for likes as a whole work
Recently I made a large system on sphinx, about 16 million records, it pulls about 100 requests per second fulltextserch