A Container-Based I/O for Virtual Routers: Experimental and Analytical Evaluations

Running repeatable and controlled virtual routing experiments

“…In some cases, this even leads to conclusions about the applicability of hypervisor‐based VMs for data plane virtualisation . The evolution of Xen's I/O performance through releases has also been analysed , and it has been shown that some of the Xen's limitations can be overcome with simple modifications . This paper also shows (although not using Xen but Kernel‐based Virtual Machine (KVM)) that much higher performance can be reached in the data‐plane of hypervisor‐based VMs than the 50–70 Kpps limits shown in some previous work.…”

“…This mechanism, called ‘NAPI‐macvlan’ is shown to stabilise routing performance in overload situations. Another way to improve the performance of the packet‐passing mechanism could be to group packets together and convey them in batches to reduce the I/O overhead , or to directly map the network card buffers in user space (in the VMM address space). This approach tries to reduce the overhead introduced by the OS kernel and allows the VMM to use batching, data prefetching and other optimisation techniques.…”

A novel methodology for efficient throughput evaluation in virtualized routers

“…The virtualization method affects significantly the I/O devices' performance, as it has been widely studied for both hardware and software virtualization [15]. For example, access to the hardware is limited to the host kernel at hardware-level, receiving requests from the virtual Network Interface Controller (vNIC) driver in the guest kernel.…”