This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: End-System Optimizations for High-Speed TCP Jeffrey S. Chase, Andrew J. Gallatin, and Kenneth G. Yocum Department of Computer Science Duke University Durham, NC 27708-0129 chase, grant @cs.duke.edu, [email protected] Abstract Delivered TCP performance on high-speed networks is often limited by the sending and receiving hosts, rather than by the network hardware or the TCP protocol im- plementation itself. In this case, systems can achieve higher bandwidth by reducing host overheads through a variety of optimizations above and below the TCP proto- col stack, given support from the network interface. This paper surveys the most important of these optimizations and illustrates their effects quantitatively with empirical results from an experimental network delivering up to two gigabits per second of end-to-end TCP bandwidth. 1 Introduction Modern TCP/IP implementations can transfer data at a high percentage of available network link bandwidth, re- flecting the success of many years of refinements. On the fastest networks, end-to-end (application-to-application) throughput is often limited by the capability of the end systems to generate, transmit, receive, and process the data at network speeds. Delivered performance is deter- mined by a combination of factors relating to the host hardware, interactions between the host and the network adapter, and host system software. This paper explores the end-system factors that can limit bandwidth for TCP on high-speed networks and the techniques to overcome those limitations. It is tempt- ing to suppose that the advances in CPU power given by Moore’s Law will render these limitations increasingly irrelevant, but this is not the case. The limiting factor is not CPU processing power but the ability to move data through the host I/O system and memory. Wider data- paths can improve raw hardware bandwidth, but more bandwidth is invariably more expensive for a given level of technology. Advances in network bandwidth follow This work is supported by the National Science Foundation (EIA- 9870724 and EIA-9972879), Intel Corporation, and Myricom. a step function, but the fastest networks tend to stay close to the limits of the hosts. Given a sufficiently fast network, achievable TCP performance depends on opti- mizations to minimize networking overheads. With Gi- gabit Ethernet widely deployed and 10 Gb/s Ethernet on the horizon, these optimizations are highly relevant to- day. This paper focuses on approaches to low-overhead networking that are now emerging into common prac- tice. The key techniques discussed are interrupt coalesc- ing, checksum offloading, and zero-copy data movement by page remapping. We also explore the effect of larger packet sizes, e.g., the Jumbo Frames standard for Ether- net. The network interface plays a key role for each of these features, and an increasing number of commercial network adapters support them. We quantify their per- formance potential and briefly outline alternative struc-...
View Full Document
This note was uploaded on 01/07/2011 for the course CSE 100 taught by Professor Staff during the Spring '08 term at UCSD.
- Spring '08
- Computer Science