Aerospike Database delivers consistent, predictable high performance at low TCO. Aerospike has been benchmarked many times. In the latest benchmarks, we show 1 Mtps on a single node and the behavior of a database cluster as new nodes are added and removed.

Server Specs

For this demo, we used the following hardware: Aerospike DB Host

  • CentOS 6.5 (2.6.32-431.3.1.el6.x86_64)
  • 2 x Intel Xeon X5690@3.47 GHz
  • 48 GB DDR3 1333 MHz RAM
  • 10 Gb Network
  • 1 x 500 GB 7200 rpm hard disk
  • No changes were made from the default operating system, other than running the Aerospike afterburner script (available at /opt/aerospike/bin/afterburner.sh)

Client Hosts (4 hosts) 4 hosts are necessary since each host was able to generate 260 ktp,

  • CentOS 6.3 (2.6.32-358.6.1.el6.x86_64)
  • 2 x Intel Xeon E5-2665@2.4 GHz
  • 24 GB DDR3 1333 MHz RAM
  • 10 Gb Network
  • 1 x 300 GB 7200 rpm hard disk
  • The afterburner script was also used here to balance network interrupts across cores.

Benchmark Parameters

For the video above the software used was the Aerospike Java Benchmark tool, which is included with the Aerospike Java Client v 3.0.22. The Java JVM used was the Oracle JDK 1.7.0_25-b15. The following parameters were used:

  • 1,000,000 keys
  • keys and records were strings of length 8
  • 96% read/4% write

Running the Benchmark

In order to load the data, run the following command from any one of the client hosts:

./run_benchmarks -h [hostname] -p 3000 -n test -k 1000000 -l 8 -o I -w I -z 64

In order to run the client load on each client host, run the following:

./run_benchmarks -h [hostname] -p 3000 -n test -k 1000000 -l 8 -o I -w RU,96 -z 64 -g 260000

Special Note

Aerospike has also done these tests using the YCSB and still maintained more than 1 million tps. If you want to recreate these, please look at the benchmark paper released by Thumbtack. This includes their configurations.