April 27, 2006
Measuring MySQL Server Performance for Sensor Data Stream Processing
Jacom Nikom (MIT Lincoln Lab and Boston MySQL Meetup regular) is presenting about performance when processing sensor data stream.
Where does the data come from? Multiple locations, 10 pictures of different kinds of sattelite and telescope equipment. I get the sense that there is a ton of data that is being collected. Sensors send messages to a publish/subscribe middleware which apply algorithms, sends it through a historian and down to MySQL.
The historian creates a thread for each subscription. The historian processes the message and creates an INSERT query. This is happening from many sensors, each operating at as many as 100 messages/second.
When choosing a database they looked at benchmarks, tried to determine performance based on the predicted databases. Wanted to know how MySQL would respond to moving.
Jacob gets into some serious detail about performance on various versions of Linux, hardware, and MySQL. The slides are the better place to get this info. This is one of the most in-depth performance reviews I've ever seen, lots of variables and plots.
Of note is a look at comparison between 4.1 and 5.0. MySQL 4.1 outperforms 5.0 in all record lengths. MySQL 4.0 is slower than 5.0 for both small and large records. So MySQL 4.1 was the best performance for their application.
Posted by mike at April 27, 2006 2:24 PM