TCP/ATM experiences in the MAGIC testbed

Abstract

This paper describes performance measurements taken in the MAGIC gigabit testbed relating to the performance of TCP in wide area ATM networks. The behavior of TCP with and without cell level pacing is studied. In particular, we focus on results that indicate that the TCP rate control mechanism alone is inadequate for congestion avoidance and control in wide-area gigabit networks. We also present results showing that TCP augmented by cell-level pacing addresses these problems and allows the full bandwidth capacity to be utilized. These results demonstrate the viability of high performance distributed systems based on wide area ATM networks given the proper ATM traffrc management infrastructure. 1: IntroductionThe behavior of the Transmission Control Protocol (TCP) [9J has been studied by a variety of authors [3,5,7,11,12, 161. Understanding TCP performance in networks based on SONET and ATM is of fundamental importance to the evolution of the existing national infrastructure to higher capacities, and hence to the growth of high performance distributed computing in wide area networks. This paper presents results from experiments conducted on the MAGIC gigabit testbed. During the initial testing of the MAGIC terrain visualization application [lo], observed throughput was less than that expected given the link capacity, processing capabilities of the hosts involved, and individual host to host throughputtests. The experiments we report in this paper were performed in an attempt to explain these anomalies and find solutions to the throughput limitations. We focus on results that indicate that the TCP rate control mechanism alone may be inadequate for congestion avoidance and control in wide-area gigabit networks. We also present results showing that TCP augmented by celllevel pacing addresses these problems and allows the full bandwidth capacity to be utilized. 1.1: Overview of ResultsThe default TCP/IP performance over congested ATM networks is poor. These congestion and buffer overflow conditions, which are caused by bandwidth mismatches or mulThe authors can be contacted via tiple sources contending for the same link, are extremely common in data networks with high performance. A number of simulation studies [5, 111 have predicted this behavior, and the measurements from MAGIC confirm this. Simulations [5] as well as preliminary experiments at the Digital Equipment Corporation Systems Research Center and in MAGIC indicate that strict link-level ATM flow control will avoid congestion-induced cell loss and provide excellent TCP throughput in local area networks, although this may be a extremely expensive solution in wide area networks. The experimental results from MAGIC address several issues. First, is TCP rate control effective in wide area ATM networks as cumently implemented? Second, if TCP rate control is not adequate, why is this the case? Third, can solutions at the ATM cell level, in particular tr&c pacing, be used to improve performance? 2: Background 2.1: MAGIC NetworkThe Multidimensional Applications and...

Performance analysis in high-speed wide area IP-over-ATM network: top-to-bottom end-to-end monitoring

“…The minimum transmission unit (MTU) size, TCP window size, and TCP write buffer size are some of the parameters that need to be tuned to optimize performance in an ATM network. These issues have been investigated by our colleagues a t t h e University of Kansas and Minnesota Supercomputer Center, and are described in detail in [12,13,141. The pacing of ATM cells (bandwidth-limiting) out of fast workstations such as the DEC 3000/600 is also very important, and is described in detail in [14].…”

“…(These experiments will be run in the near future, and we will post the results t o http://wwwitg.lbl.gov/DPSS/Experiments.) Experiments done at the University of Kansas, Lawrence, show t h a t both large MTU size (illustrating the problem noted above) and small MTU size (preventing high-performance operation of the servers) lead to low throughput [13]. Figure 10 (data from the KU experiments done in the MAGIC testbed) indicates the relationship between TCP throughput with varying MTU size for the kind of large windows (196 kbytes) used in a high-bandwidth network.…”

The MAGlC project: from vision to reality

“…Experiments showed that throughput close to the maximum theoretically possible could be attained on OC-3 links over long distances. To achieve high throughput, both the maximum transmission unit (MTU) and the Transmission Control Protocol (TCP) window must be large, and flow control must be used to ensure fairness and to avoid cell loss if there are interacting traffic patterns [5,6].…”

Robustness-Agile Encryptor for ATM Networks

“…Catch-phrases like "broadband to the curb" or "broadband to the home" characterize a trend towards ubiquitous high-speed telecommunication [1] [2]. Asynchronous transfer mode {ATM) is a technology of choice and has matured from first pilot installations and research test beds in the early 1990s [3] to continuously increasing popularity as demonstrated in its use in the Trans European Network [4]. The benefits promised by ATM can be summarized by sketching two major characteristics: First, ATM is a technology that can deliver different types of traffic with Quality of Service (QoS) over a single digital transport mechanism.…”