QVISOR: Virtualizing Packet Scheduling Policies

Abstract

The concept of programmable packet scheduling has been recently introduced, enabling the programming of scheduling algorithms into existing data planes without requiring new hardware designs. Notably, several programmable schedulers have been proposed, which are capable of running directly on existing commodity switches. Unfortunately, though, their focus has been limited to single-tenant traffic scheduling: i.e., scheduling all incoming traffic following one single scheduling policy (e.g., pFabric to minimize flow completion times). In this paper, we emphasize the fact that today’s networks are heterogeneous: they are shared by multiple tenants, who run applications with different performance requirements. As such, we introduce a new research challenge: how can we extend scheduling programmability to multi-tenant policies? We envision QVISOR, a scheduling hypervisor that enables multi-tenant programmable scheduling on existing switches. With QVISOR, tenants program the scheduling policies for their traffic flows; operators define how tenants should share the available resources; and QVISOR does the rest: deploying the scheduling policies into the underlying hardware.