Performance Instrumentation for Global Bus Systems
Gerard M. Gordon · 1986
Buses are conmlon computer structures. Nevertheless, high performance Global Bus Systems are difficult to design, build and verify. A comaon mistake is to underestimate the need and complexity of empirical system measurement required to understand, confirm or improve early designs. Instead, ad hoc metrics and measurement techniques are frequently used that can be inaccurate or inappropriate. GBS instrumentation techniques have no formal foundations, instead they are knowledge-intensive; and perhaps worse yet, this knowledge has not been codified, so that GBS designers are each forced to re-discover it as best they can. This report adds structure to the field of GBS instrumentation by defining a set of metrics and data collection techniques important in serial global bus system analysis. Emphasis is placed at the physical and data link levels which correspond to the lower levels of the ISO Open Systems Interconnection (OSI) reference model and the IEEE Std 802 family of local network standards. The metrics are measures of performance, reliability, stability and correctness. Instrumentation guidelines are given depending on the location of data to monitor, required accuracy and time of instrumentation design; software, hardware and hybrid monitors, sample and event driven techniques are covered. A high level design for a network instrumentation center with these ideas incorporated is also presented. The design includes special monitoring hardware needed per bus and bus interface, and a scheme for central collection of monitored data to aid post collection analysis. A case study of a network file system where these instrumentation techniques are used appears at the end.