Accelerating packets reception and forwarding in virtualized datacenters

Abstract

In virtualized data centers, the last hop of the network is no longer handled by a physical network switch, but rather is performed in software inside the server to switch among virtual machines hosted by that server. However, the current I/O network virtualization based on the driver domain is an attractive model since it offers high flexibility and cost-effectiveness. However, it exhibits poor networking performance due to multiple memory transactions between the driver domain and the VM. This paper addresses this challenge by introducing the concept and design of a new I/O subsystem called Next Hop Bridge which identifies the packets belonging to the same flow and processes only the first one of a given flow within the virtual machine. NHB uses a flow table to store the needed information about the packets. Such a switching mechanism allows to considerably avoid unnecessary I/O transactions during the communication between the driver domain and the VM and hence make the VM throughput scale up to line rates.