Benchmarks

With the many NoSQL databases available including Couchbase, MongoDB and Cassandra, it can be difficult to decide which one is right for your application, partially because the features differ between each of them, and partially because there is not an easy way to compare the performance of one database versus another.

Many of these benchmarks were measured using the Yahoo! Cloud Serving Benchmark (YCSB), which is an independent framework and common set of workloads for evaluating the performance of different databases. You can obtain the latest Aerospike YCSB plug in at the Aerospike Github repository.


10x faster performance

Ultra-High Performance NoSQL Benchmarking: Analyzing Durability and Performance Tradeoffs

In an independent benchmark test of NoSQL databases conducted by Thumbtack Technology, Aerospike wiped out the competition, delivering nearly 10 times the throughput.  The flash drive/SSD optimizations of Aerospike delivered an unprecedented speed advantage over other databases!
Read More about “Ultra-High Performance NoSQL Benchmark
In early 2014, we doubled our performance on the same YCSB tests.

Read the 1 M TPS post.


Aerospike Soars in the Amazon Cloud

Aerospike Soars in the Amazon Cloud

This benchmark demonstrates 1 M TPS in pure RAM on a single Amazon Web Services (AWS) instance for $1.68/hour. It documents experiments with a number of configurations and steps taken to achieve this high performance. It also compares read/write throughput and price/performance on several Amazon instances with a 4-node Aerospike in RAM cluster. The tests reveal the Amazon r3.large and r3.2xlarge instances result in the highest throughput across all workloads.

Read more and download the benchmark report

You can also look at the instructions on recreating this benchmark


Intel

Higher Scalability at Lower Cost for Real-Time, Big Data Applications

Aerospike continues to push the boundaries of even next generation hardware. In this paper from Intel, Aerospike DB is able to achieve a 56% improvement in transaction rate using a Xeon E5 v3 processor (as compared to the v2). Performance has been increased to a very lofty 2.5 M TPS using RAM.

Read more and download the Intel whitepaper


Cloud Spectator

Cloud Spectator Whitepaper: Benchmarks of Bare-Metal and Virtual Cloud

To quantify the performance differential of running Aerospike’s in-memory NoSQL database on bare-metal and virtual cloud infrastructure, Cloud Spectator conducted throughput speed and latency benchmark tests for three different types of workloads on Internap’s bare-metal servers and AWS EC2 I2 class and Rackspace Performance Cloud Servers virtual instances. Key findings include:

  • Loading Data Workload – This analysis tested “write” speed when loading data into the database, similar to new users populating their account information into a website. Internap outperformed Rackspace by 5x and Amazon by 51 percent on throughput speed. Internap had 77 percent less latency than Rackspace and 56 percent less latency than Amazon.
  • Balanced Workload – This analysis tested a balance of 50 percent database “reads” and 50 percent updates, simulating the activity of an active ecommerce user and corresponding personalization of digital advertising based on that activity. On throughput speed, Internap outperformed Rackspace by 2.7x and Amazon by 50 percent. Internap had 59 percent less latency than Amazon and 32 percent less latency than Rackspace.
  • Read-Heavy Workload – The read-intensive workload, which is associated with applications like content tagging, featured 95 percent “reads” and 5 percent updates. On throughput speed, Internap outperformed Rackspace by 2.5x and Amazon by 61 percent. Internap had 51 percent less latency than Amazon and 48 percent less latency than Rackspace

Download the full report


Failover Characteristics of Leading NoSQL Databases

Failover Characteristics of leading NoSQL databases

In this study, Thumbtack Technology examines one of the main reasons for using a NoSQL database — the ability to continue processing transactions in the face of hardware or other node failures. Instead of just focusing on raw performance numbers, it also explains how failure and recovery events affect the system as a whole.

Get the “Failover Characteristics Benchmark” paper