Collaborative Hardware Design for Real-Time Processing

• Activity Number: 33
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Computing
• Author(s): Sarah Michalak*+ and Andrew DuBois and David DuBois and Christine Anderson-Cook and Stephen Poole
• Companies: Los Alamos National Labs and Los Alamos National Laboratory and Los Alamos National Laboratory and Los Alamos National Laboratory and Oak Ridge National Laboratory
• Keywords: fault injection ; real-time processing ; soft error

Abstract

Designing hardware for real-time processing presents a challenging constraint space. A fault injector for I/O (FIFI) was developed to enable real-time fault injection into I/O packets so that the effects of such faults could be carefully investigated. As such, FIFI's design and capabilities had to meet certain requirements deriving from the need to use it in statistically designed experiments. Statisticians worked jointly with the hardware engineers who designed FIFI to ensure that these requirements were met.

This talk presents some of these requirements and details how they were addressed in FIFI's design given the hard constraint of essentially real-time fault injection. The trade-off space between implementing features in hardware versus software is discussed.