Automated Workload Characterization for I/O Performance Analysis in Virtualized Environments
Proceedings of the 6th ACM/SPEC International Conference on Performance Engineering - ICPE '15
Modern applications, such as mail servers, file servers, or video servers showhighly I/Ointensive workload patterns. Their huge data volumes require powerful storage infrastructures. These applicarions are increasingly deployed in virtualized environments due to cost efficiencyaspects. Nevertheless, consolidating several applications on one shared infrastructure introduces complexperformance implications due to mutual interferences. To consolidate several applications while respecting certain
... especting certain Service LevelA greements necessitates apredicion of these implications up front. Such aprediction, however, requires tailored performance models that in turn require asignificant amount of expertise to create the models [Kr12, Kr11]. Moreover, their accuracyd epends on the quality of the input parameters which are often unclear howtheycould be determined [CH11]. We adress this discrepancyi no ur work. We develop an automated workload characterization approach to extract workload models [Ko09] that are representations of the main aspects of I/Ointensive applications in virtualized environments. We have tailored our approach to enable anon-invasive and lightweight monitoring, yet with alevel of abstraction such that the parameters are practically obtainable. To evaluate our approach, we perform acomprehensive evaluation demonstrating its workload modeling performance for common business workloads using twocase studies. The case studies demonstrate typical real-worl scenarios, such as consolidation of several workloads on one machine, and workload migration between twosystems.