Many of the software systems we use for everyday purposes incorporate elements developed and maintained by third parties. These elements include not only code components and data bases but also dynamic data feeds from online data sources.? Even though everyday software is not mission critical it must be dependable enough for its intended use. This is limited by the dependability of its constituting elements.

It is especially difficult to assess the dependability of dynamic data feeds because they exhibit not only "fail-silent" behavior but also semantic failures -- delivery of unreasonable yet well structured results by a responsive data feed.? Further, it is normal for the behavior of such data feeds to change.? Unfortunately, the specifications of these data feeds are often too incomplete and sketchy to support failure detection.

We propose an approach for benchmarking the semantic availability of redundant data feeds. The fault model is defined as violations of inferred invariants about the usual behavior of a data feed.

Preferred citation: Orna Raz, Philip Koopman, and Mary Shaw. Benchmarking semantic availability of dynamic data feeds, Workshop on Dependability Benchmarking (in conjuction with DSN), 2002.