|
Abstract:
|
In datacenters the availability of servers and other equipment is routinely monitored by recording uptime to total time. Uptime is based on the ability of the applications, machines, or the entire facility to provide an acceptable level of service and is usually defined as a percentage over a time window, such as a week or a month, plotted in calendar time. The cumulative uptime percentage is also used. However, such summary statistics can be poor measures, since they do not account for time dependence of failures and multicensoring. Furthermore, since availability is a function of both outage frequency and duration, there is no distinction between large numbers of small outages and small numbers of large outages. This paper describes approaches based on extensions to Mean Cumulative Functions used successfully at Sun Microsystems to analyze the reliability of repairable systems.
|
