Performance analysis of IPv4 vs. IPv6 on various operating systems using jumbo frames

Abstract

The services offered on the Internet, the requirements and demands for use of multimedia applications, such as sharing video files, photos, video conferencing, and the increased usage of social networking sites, have all contributed to the exponential growth of the Internet usage over the last ten years. In order to meet these demands, hardware developers have increased the speed of hardware such as processors and switches, and routers, also have increased capacity of the communication medium. However, the maximum data unit that can be passed onwards by any layer via this communication medium remains untouched. The current Maximum Transfer Unit (MTU) is 1500 Bytes, to address this issue, a new MTU has been emerged which is known as Jumbo frames. The proposed MTU for Jumbo frames in now 9000 Bytes.

The purpose of this study is to evaluate the performance of Jumbo frames on a network environment employing six operating systems from two different distributions. These operating systems are; Microsoft Windows Server 2008, Microsoft Windows Server 2003 and Microsoft Windows 7 Professional and from the Linux distributions, Linux Fedora, Ubuntu and OpenSUSE. In this study, two transmission protocols were employed namely, Transmission Control Protocol (TCP) and the User Datagram Protocol (UDP). Two Internet protocols were also engaged in these performance experiments, Internet Protocol version six (IPv6) and Internet Protocol version four (IPv4). There were five main performance metrics extracted from the data collected in this experimental study namely the throughput, delay, jitter, the CPU utilisations on the software routers and the packets dropped rate. The experiments were conducted on the network level and at the application level with the following applications; DNS, Games and VoIP traffics on five different VoIP CODECs. The Jumbo frame sizes involved ranging from 1518 Bytes to 9014 Bytes. The findings of this study concluded that for traffic employing TCP as transport protocol, Microsoft Windows Server 2008 and Microsoft Windows 7 yielded the highest throughput on both IPv6 and IPv4 and also Linux OpenSUSE on IPv4 only. When UDP was employed as transmission protocol, all of the operating systems yielded similar throughput values

With regards to the applications’ results, the findings of this study concluded that for DNS, Microsoft Windows Server 2008 and Microsoft Windows Server 2003 yielded the highest throughput on IPv6 and Linux Fedora on IPv4. For the games on IPv6, Linux Ubuntu yielded the highest throughput on the CSa game while all of the operating systems yielded similar values on the CSi game. On the Quake3 game, Microsoft Windows Server 2008 produced the highest throughput values. iii

With regards to the VoIP on IPv6, Microsoft Windows Server 2003 and Linux Ubuntu yielded the highest throughput on the G.711.1 CODEC while OpenSUSE yielded the highest values on the G.711.2 CODEC hence; Microsoft Windows Server 2008 generated the highest amount of delay on all of the CODECs. With regards to VoIP on IPv4, Microsoft Windows 7 yielded the highest throughput on the G.711.1 CODEC while Microsoft Windows 7, Linux Fedora and OpenSUSE yielded the highest values on the G.711.2 CODEC however, on both IPv6 and IPv4 for the G.723.1, G.729.2 and the G.729.3 CODECs, all of the operating systems yielded similar throughput values. The good combination for IPv6 on VoIP would be Microsoft Windows Server 2003 or Linux Ubuntu on the G.711.1 CODEC, they both yielded high throughput but low delay and low jitter and for IPv4, Microsoft Windows 7 on the G.711.1 or G.711.2 CODECs; it yielded high throughput but low delay and low jitter.