PRISMA: Efficient Algorithms and Methods for Online Extraction of Performance Models in Virtualized Environments (DFG) (08/2016-07/2019)

Description: Modern IT systems have increasingly complex layered architectures composed of loosely-coupled components deployed in virtualized environments. The use of virtualization provides increased flexibility and efficiency by enabling the sharing of IT resources among multiple independent applications. However, virtualization comes at the cost of increased system complexity and dynamics. Managing the end-to-end performance of applications in virtualized environments while ensuring efficient resource usage is a challenge.

Questions such as “How quickly and at what granularity (e.g., vCores, virtual machine instances, physical machines, server clusters) should resources be allocated/deallocated to applications as workloads change?” arise during the operation. Thus, this project tries to bring models closer to the real system by composing and calibrating them at run-time based on monitoring in the target production environment using real-life operational workloads.

People involved: Samuel Kounev, Simon Eismann, Johannes Grohmann

Project end: July 2019