1. Introduction

Everybody wants their distributed applications to use a fast, efficient form of messaging. After all, if your computer had a "fast/slow" switch, would you ever set it to "slow"? (Don't laugh!) But there are some applications that need high performance messaging to be successful. These applications generally emphasize one of two different measures of performance: throughput (total bits per second that can be sustained) and latency (elapsed time between the sending and subsequent reception of a message). In the latter case, even an extra millisecond of latency can be a serious blow for these latency-sensitive applications.

This white paper is intended to help designers and implementers of latency-sensitive systems minimize their messaging latency.

There are many techniques that can be used to reduce message latency. However, as you will see, some techniques can have negative side-effects on other characteristics of a system. For example, sending smaller packets can reduce latency but can also result in greater CPU and network load. These sorts of tradeoffs are inevitable and need to be taken into account when designing a system, choosing hardware, and allocating resources. Ideally, a messaging system allows not only the selection of many of these latency-reducing techniques, but also allows fine-grained adjustment of their operating parameters so that the tradeoffs can be optimally balanced for a given system.