ScaleBase Benchmark
Even in the world of NewSQL, many vendors can’t run the standard TPCC benchmark test. This is why it’s so important for us at ScaleBase to run a benchmark similar to the TPCC test (we used the DBT-2 test, which is a fair usage implementation of TPC-C), and to show the excellent results users can expect when using ScaleBase.
We decided to use Amazon EC2 RDS as a platform for our current benchmark . With so many customers using the EC2 platform, and so many bad reviews on database performance with regard to EC2, we decided it would make an excellent case study for ScaleBase performance.
Architecture
As can be seen from the above diagram, we used 4 Extra Large EC2 machines, running the TPCC client application (we used the DBT-2 application, which you can find here) and the ScaleBase instance.
As a database, we used 16 RDS Extra Large instances, all with the TPCC schema.
Sharding was performed using Warehouse_ID, and all tables in the schema were marked as sharded.
Database Benchmark Results
There are several interesting parameters for results. The first is the actual performance, measured in NOTPM – New Order Transactions per Minute. NOTPM accounts for about a third of the overall transactions the test runs, but this is the bar for comparison.
The second is the price per transaction. Naturally, on an EC2 environment, this is difficult to calculate. We decided to check the price of the environment for a year, and use this number as the cost basis for our calculation.
So, without any further delay – here are ScaleBase RDS benchmark results.
We used 30 warehouses on every shard, which is an Extra Large RDS instance.
|
# shards |
speed TPM |
# conn |
# warehouses |
|
1 |
18670 |
100 |
30 |
|
16 |
179300 |
1000 |
480 |
So, overall, a database size of around 500GB, and 180K NOTPM. Rather good numbers, especially when considering the fact that the benchmark is write-intensive and runs on a cloud environment.
But what about pricing? Calculation shows a rate of 50 cents per transaction (meaning, roughly, a 90K machine is needed to run 180K transactions a minute, every minute, for a year) Extremely low. Do note that calculation was done on reserved RDS instances, calculated for a year, and includes the list price for ScaleBase.
Looking at these numbers, it’s easy to see that with ScaleBase, RDS instances can easily support over 9,000 transactions per second (reads and writes, calculating not just NOTPM, but the complete number of transactions). This number should easily be sufficient for even the most active web applications out there – at a great price. And if you need more, here’s the graph that shows you just how well you can scale with ScaleBase; growth is linear.
Please click here to learn more.
