While I was visiting the Concurix team getting a demo of some of the awesomeness they have for node.js profiling, Bryce Baril mentioned that he was a new node-redis core-committer and how performance was really critical for the redis package (and redis in general).
I asked what he used to benchmark and he informed me that redis has a very robust benchmarking tool built in.
I immediatley thought of using
redis-benchmark to illustrate the difference
between our (soon to be deprecated) standard flavors and our new performance
flavors. Based on some reading on the redis-benchmark page, I decided to use
the same critiera as the Redis Benchmarks page for
evaluating bare-metal and virtual machines.
> redis-benchmark -r 1000000 -n 2000000 -t get,set,lpush,lpop -q -P 16 > redis-benchmark -r 1000000 -n 2000000 -t get,set,lpush,lpop -q
In both cases, I used 1gb virtual machines, running
Ubuntu 13.04 (Raring Ringtail) (PVHVM beta) in the ORD datacenter. Redis is
2.6.7 and I ran the test 5 times on each machine.
|Pipelined (16)||No Pipeline|
I'm not claiming to be an expert on Redis performance, but it's obvious that our new performance flavors kick ass. We're seeing 35-45% faster results in the non-pipelined case, and 16-31% in the pipelined case. Not bad considering we made the pricing even more competitive.
If you haven't already signed up for our Developer Trial you should to give the new servers a spin, hopefully these numbers will help entice you!