How the “Exasol Beast” is eating up the TPC-H benchmark and scaring the living daylights out of our competitors.
We have recently completed a new attempt at the TPC-H benchmark and the results are quite startling, even by our high standards from three years ago when we set the bar already way beyond the reach of our competitors.
We are still in the process of having our 2014 results audited – and as I pointed out in a previous blog article, anybody can boast about amazing speed, but unless it is independently audited, it’s all just hot air.
So, I won’t disclose any details, except to say that (pending the audit) we’re faster than our 2011 attempts, do more with fewer nodes, and scale further.
Do more with less
The most startling result is in the 100 TB category, currently the largest recognised by the TPC organisation. There has only ever been one attempt at this benchmark, and that was a team from Hitachi who last year ran the benchmark on a system costing $ 15m and achieved over 80,000 performance units. I can’t yet disclose our performance figures, except to say that we were orders of magnitude better on a machine costing a small fraction of the price.
The cluster we used for the benchmark is known internally as “The Beast” until we come up with a better name and, for the 100 TB benchmark it comprised 50 Dell server with nearly 40 TB of memory. Only 3 years ago we required 60 servers to perform “only” the 10 TB benchmark.
This “doing more with less” indicates that our intelligent in-memory approach has become a genuine possibility for increasingly huge database sizes. Our TPC-H results show a reassuring straight-line that proves that 100 TB is by no means our limit – we could go way beyond 100 TB just by building a bigger “Beast”.
Of course, not everyone needs to run a 100 TB database and not everyone can afford to buy a “Beast” with 50 nodes – which is why we run the 100 GB benchmark on a 6 node “baby beast” to demonstrate that interstellar performance is possible on very down-to-earth equipment.
A 100 TB use case in the real-world
Additionally, because in our architecture not all of the database needs to be in memory all of the time, I know of one real-world Exasol customer whose database is bigger than 100 TB and who is currently running quite happily on “only” 16 nodes. It’s a question of trading off cost against performance, and Exasol gives you maximum flexibility to choose.
I look forward to discussing the results in more detail when they are available, hopefully later on this month. Prepare to be amazed, and if you have a good name for the Exasol “beast”, please do let us know.