High Availability in IT Systems: How Redundancy Reduces Downtime and Adds the Nines

Abstract

ABSTRACT Information systems are expected to be available whenever they are needed, but the components they run on, the servers, disks, and links, fail from time to time, so a system built on a single component is unavailable whenever that component fails; redundancy, providing more than one component so that the system keeps working if one fails, is the principal way to make a system highly available. This study examines high availability in IT systems and how redundancy reduces downtime, the availability of a system being conventionally described by its number of nines. The availability of a system, and the downtime it suffers in a year, were examined as redundant components were added, each component being available most but not all of the time, so that the effect of redundancy on availability could be seen. Adding redundant components raised the availability sharply, each one cutting the downtime by a large factor: a single component, available most of the time, still suffered substantial downtime over a year, but with a second component to take over when the first failed, the system was unavailable only when both failed at once, which was far rarer, so the downtime fell greatly, and a third component reduced it further still. Each redundant component thus added roughly another nine to the availability, taking it from, say, ninety-nine per cent toward ninety-nine point nine nine and beyond, and cutting the yearly downtime from days to minutes to seconds. There were diminishing returns and rising cost, however, since each further nine required as much again in redundancy for an ever smaller absolute gain, so very high availability is expensive. The study shows that redundancy reduces downtime by large factors, adding nines to availability, and it discusses the implications, and the costs, of building highly available IT systems.

KEYWORDS: High availability, Redundancy, Downtime, Fault tolerance, Reliability, The nines, IT systems, Dependability, Single point of failure