Introduction

Network performance of computer systems is important to many organizations, especially in this era of the Internet. Therefore, Linux network developers have been working hard to improve the network stack performance of the Linux kernel. Because network performance also plays an important role in making Linux ready for the enterprise market segment, it is important to evaluate the performance progress of the Linux network stack to determine how well Linux handles various network workloads.

This chapter presents a case study that evaluates network performance on Linux through the use of various industry-standard benchmarks. Also included in this chapter is a discussion of network features whose performance was enhanced in the Linux 2.4, 2.5, and 2.6 kernels, and a discussion of the performance gain achieved by each feature, individually and cumulatively throughout the operating system.

The SUSE and Red Hat Linux distributors introduced the 2.6 Linux kernel in the SLES 9 and RHEL 4 releases. They have also both backported most of the 2.5 and 2.6 kernel features to their 2.4 kernel-based releases. Although the study presented in this chapter looks at both the 2.4 and 2.5 kernels, it also reflects the performance of the 2.6 distribution kernel as the 2.5 development kernel morphed into the 2.6 release level kernel.

Chapter 12, "Network Tuning," discusses the network parameters that can be tuned to increase the network's efficiency and performance. This chapter shows how network tuning was done in a case study to establish the baseline from which performance gains were measured.

Because TCP/IP is the most commonly used Internet protocol, this chapter concentrates only on the TCP/IP stack and Ethernet network. The terms Linux TCP/IP stack and Linux network stack are used interchangeably because the TCP/IP stack is part of the Linux kernel and is also considered the default Linux network stack.

We'll begin by briefly looking at each of the benchmarks used in the case study.