Efficient routing and reconfiguration in virtualized HPC environments with vSwitch-enabled lossless networks

摘要

To meet the demands of communication-intensive workloads in the cloud, virtual machines(VMs) should utilize lowoverhead network communication paradigms. In general, such paradigmsenable VMs to directly communicate with the hardware by means of a passthrough technologylike Single-Root I/O Virtualization (SR-IOV). However, when passthrough-based virtualization iscoupled with lossless interconnection networks, live migrations introduce scalability challengesdue to the substantial network reconfiguration overhead.With these challenges inmind, we proposeda virtual switch (vSwitch) SR-IOV architecture for InfiniBand in our previous work titled“Towards the InfiniBand SR-IOV vSwitch Architecture”. In this paper, we first suggest solutionsto rectify the space-domain scalability issues that are present in vSwitch-enabled subnets as aresult of the VMs using dedicated layer-two addresses. Then, we discuss routing strategies forvirtualized environments using vSwitches and present a routing algorithm for Fat-Trees.We alsopresent a reconfiguration method that minimizes imposed reconfiguration overhead on Fat-Trees.We perform an extensive evaluation of our prototype algorithms, and as vSwitch-enabled hardwaredoes not yet exist, we deduce from empirical observations by emulating vSwitches withexisting hardware, as well as large-scale simulations. Our results show significant reduction inthe reconfiguration times as route recalculations can be eliminated, and for certain scenarios, thenumber of reconfiguration subnetmanagement packets sent to switches is reduced from severalhundred thousand down to a single one without degrading the routing quality.