Performance benchmarking of application monitoring frameworks

Viac o knihe

Application-level monitoring of continuously operating software systems offers insights into their dynamic behavior, crucial for maintaining performance and availability during runtime. However, this monitoring can introduce significant runtime overhead, influenced by the number and placement of instrumentation probes. To enhance instrumentation and minimize monitoring overhead, understanding the performance impact of each probe is essential. While many frameworks claim minimal performance impact, these assertions often lack comprehensive evaluations of the actual monitoring costs. Benchmarks can effectively assess this impact, yet no specific benchmark focused on monitoring overhead currently exists, nor is there an established methodology for benchmark engineering that guides the design, execution, and analysis of such benchmarks. This thesis presents a benchmark approach to evaluate the performance overhead of application-level monitoring frameworks. Key contributions include a definition of common monitoring overhead causes, a general benchmark engineering methodology, the MooBench micro-benchmark for quantifying monitoring overhead, and detailed evaluations of three different monitoring frameworks. Extensive experiments validate the approach and benchmark results, which are available as open-source software to support further validation and replication.