How Fast is Fast and can we make EXASOL faster?
There’s no real debate about who is the “Fastest Man on Earth”. It’s currently Usain Bolt – we’ve seen him run the 100 Metres faster than anyone else, so there’s surely no argument.
The 100 Metres is an independent and audited benchmark. You have to run on a certain track, wear certain shoes, and you have to avoid certain drugs. You definitely aren’t allowed to use a motorcycle. If you did try to cheat, there are a bunch of people in the stadium and in the testing labs who would find you out. The viewers at home would definitely notice the motorcycle.
The TPC-H benchmark is a similar idea in high-performance database computing. You need to run a defined set of queries that are representative of “real-life” computing and, because database vendors have been known to exaggerate, there’s a stringent independent audit before you can publish the results.
The TPC-H queries start easy and get increasingly complex. Some of the later queries require the use of “correlated sub-queries” and this trips up more than a few of our competitors.
Athletes race over a range of distances from the sprint to the marathon. Similarly, high-performance databases can set TPC-H benchmarks on databases starting at 100GB and going up to 100TB.
We (EXASOL on Dell hardware) currently hold the TPC-H world records at 100GB, 300GB, 1TB, 3TB and 10TB. We set these records in 2011 and nobody has come close since. And when I say “nobody has come close”, that’s no exaggeration. In the 10TB class, our nearest competitor managed less than half a million TPC-H performance units. We managed SEVEN million.
We have been continually improving our product over the last three years – every change we make is tested, not just to check that it doesn’t break something else, but also we check that the change hasn’t slowed anything down. We have a commitment to optimizing the speed of our product that would not be out of place in the world of motor racing.
With that in mind, we think it is time to attempt the TPC-H benchmarks again – preliminary (unaudited) results indicate that we will put the records we hold even further beyond the reach of our competitors, take the currently vacant 30TB record and smash the 100TB record (currently held by Hitachi) by orders of magnitude.
We think that the TPC-H benchmarks are a good way of proving that our product has the performance and linear scalability that is needed by companies looking to mine their mountains of “Big Data” for nuggets of golden insight.
Of course, no abstract benchmark is absolute “proof” that our database would be the fastest for your business – which is why we offer a free 14-day Test Drive so that you can easily try your queries against your data using our servers.
Our competitors say that they are fast too – but talk is cheap – can they prove it over 100 metres with no cheating?