Abstract

Modern messaging, trading, and reporting systems require extremely low latencies, in the range of a few milliseconds, to meet customer expectations and strict service level agreements. This is in contrast to online commerce or other human-facing systems that can tolerate response times of a few full seconds. For most low latency systems requirements continue to get even tighter – just a fewyears ago ultra-low latency systems were those below 100 milliseconds, now they are often in the microsecond range.

The Java language never used to be strong in this space. Traditional JVMs have inherent limitations that prevent achieving ultra-low latency. Chief among them is Garbage Collection (GC). This white paper discusses the historic pitfalls of Java application latency and jitter behavior and explains how modern Java applications can achieve the long-sought-after goal of continuously consistent latency at high sustainable throughputs. The paper then describes how to use new JVM technology to achieve worst case platform latencies of 10 msec or less “out of the box” and maximum platform latencies of 1-2 milliseconds with very little tuning effort.