EXASOL has historically been seen as a mainly DACH (Germany, Austria and Switzerland) based vendor of data warehousing. Outside central Europe it has had little success until recently. However, last year it formed a partnership with Intelligent Edge Group in the UK and this has now led to the formation of a formal EXASOL presence here. Moreover, it already has British customers (most notably King, the developers of Candy Crush Saga—about which more anon). It addition, the company has opened offices in the States and Brazil and is partnering elsewhere (for example, Israel).
King is interesting for a number of reasons. The first is that it runs a large (one petabyte) Hadoop cluster alongside EXASolution (the company’s product), which was initially implemented as a four node cluster that it is now expanding. King’s initial testing showed that, for some analytical tasks, 4 EXASOL nodes were equivalent to a 420 node Hive cluster. Even if each of those nodes is pretty cheap that’s a big saving, and explains part of EXASOL’s success. Secondly, the two companies are working together to extend joint Hadoop/EXASolution capabilities. Thirdly, King is typical of EXASOL clients, to the extent that it is a tech company in its own right. One of the things about EXASolution is that it is bundled with Centos (which improves performance and simplifies maintenance, amongst other things) and, while non-technically savvy companies might look askance at this, companies like King have no problem with this approach. It is possible to run on virtualised environments or in the cloud if you prefer.
As far as the technology is concerned, EXASolution is an in-memory database that is similar in one sense to the way that IBM uses in-memory in its BLU Acceleration technology. That is, it uses as much memory as it can but has been designed to recognise that many analytic queries will need to run against very large datasets that cannot affordably be placed entirely into memory. However, EXASOL is different from most other in-memory database vendors in the sense that it was designed from the outset to be columnar and parallel and in-memory and this makes it much more efficient than other recently announced ‘add-on’ data-in-memory products.
The latest release of EXASolution is version 5.0, which was introduced last month. In addition to performance improvements it includes the provision of installable R, Python and Java packages, graph analytics capabilities (geospatial capabilities were already available) and, most importantly, the introduction of an in-database analytic tool called Skyline.
Skyline provides what is known as preference analytics, which was developed in conjunction with the University of Augsburg (EXASOL has a history of such collaborations). This applies to multi-criteria decision making. To give a simple example, you want to invest in the fund that provides the highest return and the lowest risk. Unfortunately, investment funds do not work that way so there is no right answer as to which fund to select. However, using Skyline you can tell the system that that is your preference and it will provide you with a short list of candidates——effectively eliminating all candidates that are worse in every respect than competitors. Now, you can do this using conventional SQL but it is laborious and slow. What EXASOL has done is to extend the SQL (SQL 2008) WHERE clause by allowing you to definite your preferences, making this process much simpler and faster to execute.
Finally, the other thing that EXASolution is well-known for is its domination of the TPC-H benchmarks. Until recently, when TPC changed its query profiles, EXASolution offered the best performance and best price/performance all the way up to 10TB. The company intends to rerun against the new profile and confidently expects to top the charts once again, this time at all scale factors up to 100TB. I’d be surprised if it didn’t. That makes EXASolution seriously interesting—a fast, low-cost platform with innovative technology—what’s not to like?