The Exasol DWH is a parallel RDBMS running on a cluster of Intel server nodes. The data is stored in shared-nothing manner but with redundant copies on other nodes, so no data is lost if some nodes disappear.
The high performance is achieved by using much RAM (think of 20 or more servers with 32 GB...), working with compressed data and doing as much as possible in parallel with local data. On a well sized system the query optimizer can use in-memory algorithms instead of disk centric algorithms which gives a big boost to full-table scans, sorts, complex joins. RAM access is hundreds of times faster than disk access.
The system is designed for complex queries on vast amounts of data, e.g. ROLAP, Business Intelligence and so on. For fast inserts or updates a conventional dbms would be the better choice.
On their homepage (www.exasol.com) there is an english whitepaper available for download after registration.
I think we all know that I'm not a big fan of using memory-based storage, but if that's how the whole system is designed to work and they're not promising fast updates and inserts - which I assume implies that they're not trying to officially own the data - then I'll probably get over it.
My most recent discovery, EXASOL, provides a database that is MPP, compressed and memory-based, rather than disk-based. They have also published some impressive looking TPC-H results (here and here). And that's pretty much all that I know about them.
Anybody know more about them?