Abstract

In this talk I will give a survey of solutions and tools that we have developed in recent years to achieve extremely high packet processing rates in commodity operating systems, running on bare metal and on virtual machines. Our NETMAP framework supports processing of minimum-size frames from user space at 10 Gbit/s (14.88 Mpps) with very small CPU usage. Netmap is hardware-independent, supports multiple NIC types, and it does not require IOMMU or expose critical resources (e.g. device registers) to userspace. A libpcap library running on top of netmap gives instant acceleration to pcap clients without even the need to recompile applications. VALE is a software switch using the netmap API , which delivers over 20_{Mpps per port, or 70 Gbit/s with 1500-byte packets. Originally designed to interconnect virtual machines, VALE is actually very convenient also as a testing tool and as a high speed IPC mechanism. More recently we have extended QEMU , and with a few small changes (using VALE as a switch, paravirtualizing the e1000 emulator, and with small device driver enhancements), we reached guest-to-guest communication speeds of over 1 Mpps (with socket-based clients) and 5 Mpps (with netmap-based clients). Netmap and VALE are small kernel modules, part of standard FreeBSD and also available as add-on for Linux. QEMU extensions are also available from the author and are being submitted to the qemu-devel list for inclusion in the standard distributions.}

LINKS : http://info.iet.unipi.it/luigi/netmap/ http://info.iet.unipi.it/~luigi/vale/