Stephen Pink

Small forwarding tables for fast routing lookups

For some time, the networking community has assumed that it is impossible to do IP routing lookups in software fast enough to support gigabit speeds. IP routing lookups must find the routing entry with the longest matching prefix, a task that has been thought to require hardware support at lookup frequencies of millions per second.We present a forwarding table data structure designed for quick routing lookups. Forwarding tables are small enough to fit in the cache of a conventional general purpose processor. With the table in cache, a 200 MHz Pentium Pro or a 333 MHz Alpha 21164 can perform a few million lookups per second. This means that it is feasible to do a full routing lookup for each IP packet at gigabit speeds without special hardware.The forwarding tables are very small, a large routing table with 40,000 routing entries can be compacted to a forwarding table of 150-160 Kbytes. A lookup typically requires less than 100 instructions on an Alpha, using eight memory references accessing a total of 14 bytes.

A faster UDP (user datagram protocol)

As an experiment in protocol optimizations, the authors undertook to improve the performance of a stateless protocol, namely the user datagram protocol (UDP) in the 4.3 BSD Unix kernel. The authors describe the successful optimizations that were done, along with measurements that shows a UDP performance improvement of between 25-35% on CISC and RISC systems, and overall kernal improvement of between 12% and 18%. >

Low-loss TCP/IP header compression for wireless networks

Wireless is becoming a popular way to connect mobile computers to the Internet and other networks. The bandwidth of wireless links will probably always be limited due to properties of the physical medium and regulatory limits on the use of frequencies for radio communication. Therefore, it is necessary for network protocols to utilize the available bandwidth efficiently. Headers of IP packets are growing and the bandwidth required for transmitting headers is increasing. With the coming of IPv6 the address size increases from 4 to 16 bytes and the basic IP header increases from 20 to 40 bytes. Moreover, most mobility schemes tunnel packets addressed to mobile hosts by adding an extra IP header or extra routing information, typically increasing the size of TCP/IPv4 headers to 60 bytes and TCP/IPv6 headers to 100 bytes. In this paper, we provide new header compression schemes for UDP/IP and TCP/IP protocols. We show how to reduce the size of UDP/IP headers by an order of magnitude, down to four to five bytes. Our method works over simplex links, lossy links, multi‐access links, and supports multicast communication. We also show how to generalize the most commonly used method for header compression for TCP/IPv4, developed by Jacobson, to IPv6 and multiple IP headers. The resulting scheme unfortunately reduces TCP throughput over lossy links due to unfavorable interaction with TCPs congestion control mechanisms. However, by adding two simple mechanisms the potential gain from header compression can be realized over lossy wireless networks as well as point‐to‐point modem links.

Efficient use of wireless bandwidth for multimedia applications

Two rapidly evolving and merging technologies are the Internet and wireless data communications. Another trend is the increased use of real-time multimedia applications for audio and video communication. The most popular transport protocol among these delay sensitive applications is UDP (user datagram protocol), which is a lightweight protocol that provides multiplexing among user processes and has low protocol processing overhead. UDPs checksum policy is to protect either an entire packet including UDP and IP headers or nothing in the packet at all. This conforms badly with applications that prefers errors in the payload to the loss of whole packets. Many real-time applications fall into this category. We present a new protocol, UDP Lite, that provides the kind of protection often needed by these real time applications especially when they are run over wireless networks. UDP Lite increases the flexibility of UDP by providing an optionally partial checksum. Each packet can be optionally divided into a sensitive and an insensitive part by the sender. Errors in the sensitive part will cause packets to be discarded by the UDP Lite receiver, while errors in the insensitive part are ignored by UDP Lite. We show how UDP Lite uses a wireless network more efficiently for two different scenarios. By combining UDP Lite with compressed RTP, the gain can be even higher. Simulations show that the error rate of the network can increase by almost an order of magnitude without increasing the packet loss ratio. This enables cheaper network solutions for applications such as IP telephony.

A multicast-based distributed file system for the internet

JetFile is a file system designed with multicast as its distribution mechanism. The goal is to support a large number of clients in an environment such as the Internet where hosts are attached to both high and low speed networks, sometimes over long distances. JetFile is designed for reduced reliance on servers by allowing client-to-client updates using scalable reliable multicast. Clients on high speed networks prefetch large numbers of files. On low speed networks such as wireless, special caching policies are used to decrease file access latency. The prototype implementation of JetFile is on the JetStream gigabit local area network which provides hardware support for many multicast addresses. The multicast Internet backbone (Mbone) is the wide area testbed for JetFile.

ATM as a Link in an ST-2 Internet

ATM promises to be a broadband network that provides quality of service guarantees for real time multimedia applications. ST-2 is an experimental connection-oriented protocol that provides guaranteed support for streams across an internet of hosts and routers. We describe an implementation of an ATM network as one link in an ST-2 internet. We explain how the ST-2 flowspec is mapped into an ATM hopspec during connection establishment. We then show that, after connection establishment, ST-2 header prediction can be performed on an ATM network to minimize protocol overhead. Finally, we argue that the current ATM adaption layers do not provide the right service for ST-2 or other variable length packet protocols designed to carry continuous media.

Low latency file access in a high bandwidth environment

Workstations will soon have hundreds of megabytes of main memory and will be attached to high speed networks. Many distributed applications will have to be redesigned to take advantage of these new features. Todays distributed file systems are designed for workstations with relatively small amounts of memory that are attached to medium-speed networks (e.g., Ethernets). File access latency will not decrease when todays distributed file systems are run on the newer platforms. The JetFile distributed file system is designed to take advantage of workstations and servers with large memories attached to networks where user-to-user bandwidth approaches one gigabit per second. The design of JetFile allows the high bandwidth of the network to provide low latency file access. This is accomplished by caching large data objects, i,e., directory subtrees, on reference to limit client/server communication. In addition, JetFile uses multicast extensively in its client cache consistency mechanism to help avoid costly network round trips between client and server. Reliable multicast is also used to support server replication of files in JetFile.

Multimedia and high speed networking in MultiG

Abstract This paper broadly describes some of the research activities in MultiG, a research program in multimedia and high-speed networking in Sweden, including those on multimedia applications using digital audio and video mixed with traditional applications, efficient implementation of network and transport protocols for gigabit networks (i.e., 6pm, the SICS Protocol Machine) and the MAC-level protocol, PTM, which is being designed and implemented for use on a fiber optic network.