Conferencia: “Software faults as major cause of failures in modern systems”.

Resumen: Although society is increasingly dependent on software, it is a fact that it is practically impossible to guarantee that software is perfect, due to the many complex functions that software has to perform and due to budget and time constraints of the software development process. As a result, complex software will eventually execute under faulty conditions that have not been foreseen during testing. To face this problem, it is well known, even recommended by some safety standards, that software developers adopt software fault tolerance mechanisms. Examples are masking software faults through diversity (e.g., N-version programming, recovery blocks, N self-checking programming), and detecting a wrong state of the system, in order to provide a fail-stop behavior or a degraded mode of service (e.g., concurrent error detection, checkpointing and recovery, exception handling). Since the reliability level of software system is strictly related to its behavior at runtime, in terms of manifesting faults and on failure behaviors, this seminar will introduce two approaches, which are used to analyze and to improve the system reliability.
The former is related to representativeness of fault load in software fault Injection. In order to assess the effectiveness of fault tolerance mechanisms, it becomes important to evaluate the system behavior under unforeseen faulty conditions. This can be done by deliberately injecting a fault condition into the system. This talk will present a new SFI strategy able to carefully select fault locations to achieve representative faultloads, i.e., faults which are likely to manifest during runtime.
The latter part of the talk presents some results on Software Failure Analysis techniques. These aim to characterize the dependability behavior of operational systems, and are used in many industrial sectors, such as aerospace and automotive. These techniques are valuable to engineers: for example, they allow to understand system failure modes, establish the cause of failures, prevent their occurrence, and improve the dependability of future system releases. Failure analysis is often conducted by collecting event logs (or simply logs), i.e., system-generated files reporting events of interest occurred during operations.
Speaker/Ponente: Domenico Cotroneo received his Ph.D. in 2001 from the Department of Computer Science and System Engineering at the University of Naples, Italy.