Ensuring Performance in the Virtual Environment

The advancement of virtualization technologies powers up modern cloud computing infrastructures, providing the advantages of higher degrees of resource multiplexing and of ease of management. Meanwhile, virtualized systems also pose several challenges in guaranteeing application performance, such as additional overheads and capacity variability. In this talk, I will present two solutions to ensure quality of service (QoS) in virtualized systems: (i) a novel lossless virtual network design and (ii) speculative application optimization by means of workload redundancy.

Network virtualization is particularly demanding due to the additional host processing required to forward packets and the increased packet headers needed for isolation. To tackle these performance impairments, I will present zOVN, a first-of-its-kind overlay network featuring a lossless virtual switch and a lightweight I/O processing stack. zOVN improves the end-to-end flow completion times, particularly for MapReduce-like application traffic.

While resource redundancy by increasing the virtual machine provisioning has been widely explored to guarantee QoS, its effectiveness may be hindered by its high cost and limited impact on tail performance. I will present sPARE, an approach that leverages workload redundancy as an alternative to improve performance, especially the tail. sPARE relies on a partial replication strategy combining queueing models and scheduling designs implemented in a system prototype.

I will conclude the talk with a roadmap how these two solutions are applicable to modern big-data analytics frameworks such as Apache Spark.

Robert Birke is a researcher at IBM Research Zurich Lab. He received his Ph.D. in Electronics and Communications Engineering from the Politecnico di Torino, Italy. His research interests are in the broad area of virtual resource management for large-scale datacenters, including network design, workload characterization and big-data application optimization. He has published more than 40 papers at venues related to communication and system performance, e.g., SIGCOMM, SIGMETRICS, DSN, FAST, INFOCOM, and JSAC.
Webpage