PRISM Forum - Message Replies
Topic: PRISM Questions and Answers
Topic Posted by: SRC
Date Posted: Wed Jan 12 8:33:33 US/Eastern 2000
Topic Description: Welcome to the PRISM forum! Please feel free to post your questions and comments about the PRISM assessment software here.
Posted by: Leroy Meyers
Date posted: Tue Nov 13 15:24:50 US/Eastern 2001
Subject: Making Sense of Prism
I just completed my first Prism reliability prediction. I used the same piece of equipment for which I used MIL-HDBK-217F previously. I was very interested in comparing the results, so I used a 100% duty cycle in Prism so that calendar hours would equal operating hours.
Using 217, the failure rate of the equipment is 83 failures per million hours. I inserted all the same parts into Prism, and came up with a failure rate of 6.
I'm thinking that 6 is a very low number; will anyone believe it? But then I realized that I forgot to put the software failure rate into Prism. That's understandable, because software is something that I never considered in a prediction before.
So I entered in the software parameters. That alone gave me a failure rate of 17,071!! Now I'm at the other extreme - this failure rate is way too high, I can't use this!
Has anyone else gone through this? How can I make sense of this? If you want to know the details of what I entered into Prism, you can contact me directly at firstname.lastname@example.org
Subject: Making Sense of PRISM
Reply Posted by: (email@example.com
Organization: Reliability Analysis Center (RAC)
Date Posted: Wed Nov 21 10:56:10 US/Eastern 2001
Thank you for sending us copies of your PRISM and MIL-HDBK-217 system analyses. We have reviewed these analyses and have identified that your system contains numerous Plastic Encapsulated Microcircuits (PEMs) and non-hermetic semiconductor devices operating at 100% duty cycle. When performing comparisons between PRISM and MIL-HDBK-217 one needs to keep in mind that the last update to MIL-HDBK-217 occurred in 1995. In addition, MIL-HDBK-217 microcircuit model updates last occurred in 1992 and the data that went into the development of these models was based on parts manufactured in 1991 or before (with the bulk of the data being from the 1980’s). The quality of ICs manufactured within the last 10+ years has improved exponentially. Also, it was a disputed fact that MIL-HDBK-217 was overly pessimistic in predicting the failure rates of PEMs, even when it was current. RAC data has shown (see RAC Order code: PEM2, page 84) little difference between the failure rates of PEM and hermetic integrated circuits when properly used. Other factors that have influenced your PRISM system failure rate to be lower, include failure rate improvements in other part technologies in your design and the result of your organization’s process assessments which are better than the PRISM default assessment (e.g., an average organization’s processes).
I agree that the failure rate of your software does “dwarf” the calculated hardware failure rate. The PRISM software model provides the instantaneous failure rate at the time period of interest. Your analysis identifies that failure rate after 12 months of deployment and assumes that the software is maintained to eliminate software faults. It was also identified that your system software was comprised of 134,000 lines of source code. It has also been determined that the software originated many years ago, but there are new releases every so often. This would lend me to believe that a good portion of your software code may be very mature and will not experience failures. In performing an assessment of your software, I would take this into account. For instance, if your update only addressed 1000 lines of code, you may want to perform a prediction based on only these 1000 lines. There is some risk in doing this, as interactions between the modified and unmodified code will exist. Never-the-less, I feel it would provide a better representation of the failure rate for the software in your specific application. When I run a PRISM analysis keeping all other prediction parameters constant, I obtain a failure rate of 127.4 failures per million hours for 1000 lines of code. At first glance, this still appears to be an unrealistic failure rate, however, lets put it into perspective. When converting this failure rate to the number of faults that occur in one year (100% duty cycle) I obtain a failure rate of 1 fault per year.
I hope that I have answered your issues. Don’t forget that if you have actual data for software or any other component, a PRISM user has the flexibility of using this data in their analysis or can modify the PRISM predicted value using Bayesian statistics.
If you have any questions, feel free to contact me.
Data and Information Manager
Reliability Analysis Center (RAC)
201 Mill St.
Rome, NY 13440-6916
Phone: (315) 339-7055
Fax: (315) 337-9932
RAC Web Site: http://rac.alionscience.com