Joseph D. Touch

Modeling the performance of HTTP over several transport protocols

This paper considers the interaction of HTTP with several transport protocols, including TCP, Transaction TCP, a UDP-based request-response protocol, and HTTP with persistent TCP connections. We present an analytic model for each of these protocols and use that model to evaluate network overhead carrying HTTP traffic across a variety of network characteristics. This model includes an analysis of the transient effects of TCP slow-start. We validate this model by comparing it to network packet traces measured with two protocols (HTTP and persistent HTTP) over local and wide-area networks. We show that the model is accurate within 5% of measured performance for wide-area networks, but can underestimate latency when the bandwidth is high and delay is low. We use the model to compare the connection-setup costs of these protocols, bounding the possible performance improvement. We evaluate these costs for a range of network characteristics, finding that setup optimizations are relatively unimportant for current modem, ISDN, and LAN users but can provide moderate to substantial performance improvement over high-speed WANs. We also use the model to predict performance over future network characteristics.

Dynamic Internet overlay deployment and management using the X-bone

Abstract The X-Bone dynamically deploys and manages Internet overlays to reduce configuration effort and increase network component sharing. The X-Bone discovers, configures, and monitors network resources to create overlays over existing IP networks. Overlays are useful for deploying overlapping virtual networks on shared infrastructure and for simplifying topology. The X-Bone extends current overlay management by adding dynamic resource discovery, deployment, and monitoring, and allows network components (hosts, routers) to participate simultaneously in multiple overlays. Its two-layer IP in IP tunneled overlays support existing applications and unmodified routing, multicast, and DNS services in unmodified host operating systems. This two-layer scheme uniquely supports recursive overlays, useful for fault tolerance and dynamic relocation. The X-Bone uses multicast to simplify resource discovery, and provides secure deployment as well as secure overlays. This paper presents the X-Bone architecture, and discusses its components and features, and their performance impact.

Effects of ensemble-TCP

TCP currently recalculates the state of each connection from a fixed set of initial parameters; this recalculation occurs over several round trips, during which the connection can be less than efficient. TCP control block sharing is a technique for reusing information among connections in series and aggregating it among connections in parallel. This paper explores the design space of a modified TCP stack that utilizes these two ideas, and one possible design (E-TCP) is presented in detail. E-TCP has been designed so that the network transmission behavior of group of parallel E-TCP connections closely resembles that of a single TCP/Reno connection. Simulated web accesses using HTTP/1.0 over E-TCP show a significant performance improvement compared to TCP/Reno connection bundles. This paper is first to evaluate performance using four different intra-ensemble schedulers for different workloads. In one scenario simulating a common case, E-TCP is 4-75% faster than Reno for transmitting the HTML parts of various pages, and 17-61% faster transmitting the whole pages. In the same scenario, reusing cached state speeds up repeated E-TCP page accesses by 17-53% for the HTML parts and 10-28% for the whole pages, compared to the initial access. E-TCP can also be integrated with other proposed TCP extensions (such as TCP/Vegas or TCP/SACK), to further improve performance.

Performance analysis of MD5

MD5 is an authentication algorithm proposed as the required implementation of the authentication option in IPv6. This paper presents an analysis of the speed at which MD5 can be implemented in software and hardware, and discusses whether its use interferes with high bandwidth networking. The analysis indicates that MD5 software currently runs at 85 Mbps on a 190 Mhz RISC architecture, a rate that cannot be improved more than 20-40%. Because MD5 processes the entire body of a packet, this data rate is insufficient for current high bandwidth networks, including HiPPI and FiberChannel. Further analysis indicates that a 300 Mhz custom VLSI CMOS hardware implementation of MD5 may run as fast as 256 Mbps. The hardware rate cannot support existing IPv4 data rates on high bandwidth links (800 Mbps HiPPI). The use of MD5 as the default required authentication algorithm in IPv6 should therefore be reconsidered, and an alternative should be proposed. This paper includes a brief description of the properties of such an alternative, including a sample alternate hash algorithm.

Dynamic Internet overlay deployment and management using the X-Bone

The X-Bone dynamically deploys and manages Internet overlays to reduce their configuration effort and increase network component sharing. The X-Bone discovers, configures, and monitors network resources to create overlays over existing IP networks. Overlays are useful for deploying overlapping virtual networks on a shared infrastructure and for simplifying topology. The X-Bone extends current overlay management by adding dynamic resource discovery, deployment, and monitoring and allows simultaneous participation in multiple overlays. Its two-layer IP in IP tunneled overlays support existing applications and unmodified routing, multicast, and DNS services in unmodified operating systems. This two-layer scheme uniquely supports recursive overlays, useful for fault tolerance and dynamic relocation. The X-Bone uses multicast to simplify resource discovery, and provides secure deployment as well as secure overlays. This paper presents the X-Bone architecture, and discusses its components and features, and their performance impact.

Idletime scheduling with preemption intervals

This paper presents the idletime scheduler; a generic, kernel-level mechanism for using idle resource capacity in the background without slowing down concurrent foreground use. Many operating systems fail to support transparent background use and concurrent foreground performance can decrease by 50% or more. The idletime scheduler minimizes this interference by partially relaxing the work conservation principle during preemption intervals, during which it serves no background requests even if the resource is idle. The length of preemption intervals is a controlling parameter of the scheduler: short intervals aggressively utilize idle capacity; long intervals reduce the impact of background use on foreground performance. Unlike existing approaches to establish prioritized resource use, idletime scheduling requires only localized modifications to a limited number of system schedulers. In experiments, a FreeBSD implementation for idletime network scheduling maintains over 90% of foreground TCP throughput, while allowing concurrent, high-rate UDP background flows to consume up to 80% of remaining link capacity. A FreeBSD disk scheduler implementation maintains 80% of foreground read performance, while enabling concurrent background operations to reach 70% throughput.

The TIME-WAIT state in TCP and its effect on busy servers

Hosts providing important network services such as HTTP and FTP incur a per-connection memory load from TCP that can adversely affect their connection rate and throughput. The memory requirement is directly tied to the number of connections; caching and other sharing methods will not alleviate it. We have observed HTTP throughput reductions of as much as 50% under SunOS 4.1.3 due to this loading. This paper advocates off-loading the memory requirements to the growing number of clients. This reduces server memory requirements as connection rate at that server grows due to increases in the number of clients and the bandwidth available on the network. These approaches control server memory load better with growing client load than per-transaction techniques such as persistent HTTP connections. These approaches also interoperate with persistent connections to take advantage of their other benefits. This paper describes the causes of the memory loading, called TIME-WAIT loading, and defines three methods of alleviating it that scale with increasing number of clients. We present measurements of the systems and a comparison of their properties.

The need for media access control in optical CDMA networks

Optical CDMA local area networks allow shared access to a broadcast medium. Every node on the network is assigned an optical orthogonal codeword (OOC) to transmit or receive on. OOCs are designed to be pseudo-orthogonal, i.e., the correlation (and therefore the interference) between pairs of codewords is constrained. This paper demonstrates that the use of optical CDMA does not preclude the need for a media access control (MAC) layer protocol to resolve contention for the shared media. OOCs have low spectral efficiency. As more codewords are transmitted simultaneously, the interference between codewords increases and the network throughput falls. This paper analyzes a network architecture where there is virtually no MAC layer, except for choice of the codeset, and shows that its throughput degrades and collapses under moderate to heavy load. We propose an alternate architecture called interference avoidance where nodes on the network use media access mechanisms to avoid causing interference on the line, thereby improving network throughput. Interference avoidance is analyzed and it is shown that it can provide up to 30% improvement in throughput with low delays and no throughput collapse. We validate our analysis through simulation with realistic network traffic traces.

A global X-bone for network experiments

A global Internet overlay testbed is being deployed to support the distributed, shared use of resources for network research. The Global X-Bone (GX-Bone) augments the X-Bone software system, enhancing its coordination mechanisms to support deployment of local overlays to world-wide, shared infrastructure. The GX-Bone is based on the X-Bones Virtual Internet Architecture which extends the Internet for both concurrent, parallel and recursive overlays and provides decentralized, automated deployment and management. GX-Bone supports host visualization through the NetFS file system, granting individual users compartmentalized access and control of host and router configuration and the DataRouter extension to IP loose source routing that supports application control of network-layer forwarding. GX-Bone can be installed on user-modified kernels, uniquely supporting both conventional kernel-level protocol development and coordinated global infrastructure sharing.

The RNA Metaprotocol

The recursive network architecture (RNA) explores the relationship of layering to protocol and network architecture. RNA examines the implications of using a single, tunable protocol, called a metaprotocol, for different layers of the protocol stack, reusing basic protocol operations across different protocol layers to avoid reimplementation. Its primary goal is to encourage cleaner cross-layer interaction, to support dynamic service composition, and to gain an understanding of how layering affects architecture. This paper provides a description of RNA and a recently completed initial prototype. The prototype extends the Click modular router with control capabilities including dynamic composition and discovery. These capabilities are used to demonstrate simple but flexible stacks of instances of a metaprotocol that are customizable at runtime.