NoSQL for Time Series Data Benchmark

May 26, 2010

Speed of processing is an important aspect in choosing a data storage for time series data. The fast the better.

So how fast it is when NoSQL meet time series data? The following was what I found.

Test env

Hardware/OS

  • Ubuntu 2.6.32-19-generic SMP 64bit
  • Intel(R) Core(TM)2 Duo CPU T7500 @ 2.20GHz
  • 2G memory
  • 5400RPM hard disk

we have compared:

ruby client:

  • tokyocabinet-ruby-1.30
  • mongo 1.0.1 with BSON_ext
  • redis 1.0.4

benchmark code

I pushed it to github in case you want to roll you own results.

Results

write 1M records

build index

read last 30 days ohlc by symbol

read all ohlc by symbol

Storage size

Results clearly shows Toyko Cabinet BDB was the winner.

Notes

This wasn't a full benchmark for NoSQL database, but more specific for time series data.