Skip to content

System Reliability Metrics

2 messages · Jon Stearley

#
I need to calculate some metrics such as Mean Time Between Failure  
(MTBF), etc (see http://www.cs.sandia.gov/~jrstear/ras for a more  
complete list).  I have observations like

                  start                 end                state
  1 2005-11-11 09:05:00 2005-11-11 12:20:00   Scheduled Downtime
  2 2005-11-12 13:42:00 2005-11-12 14:45:00 Unscheduled Downtime

where each row describes the system state between start and end  
times.  Time between observations indicates a Production state.  The  
metrics of interest involve simple ratios of total time spent in  
various states, number of transitions, etc.  I'd like to plot period  
MTBF values (eg monthly vertical bars), as well as a cumulative MTBF  
(eg a line whose last value indicates MTBF for the entire time  
range), etc.

What is the best approach in R towards these results?

Irregular time series seem a natural fit for the observations, and  
the reporting periods for the metrics will have regular periods - so  
I've perused the stats, tseries, and its packages.  A markov class  
could be used (eg msm package), but that's probably overkill (like  
the reliability packages concord and irr)?  My goal is to calculate  
the metrics in the simplest and most natural way.  Can someone please  
advise me on a best approach?

Thank you!
2 days later
#
On Dec 21, 2005, at 11:02 AM, Jon Stearley wrote:

            
I have found that the zoo package fills most of my needs.  Thanks.

-jon