any researchers have observed thatwhile thelinklevelratesofsome networks are now in the gigabit-persecond range, the effective throughput between remote applications is usually an order of magnitude less. A number of componentswithin computing systems have been postulated as the cause of this imbalance. Several years ago, the transport and network protocols came under heavy scrutiny, as they were considered to b e "heavyweight" a n d t h u s computationally expensive. This line of thought encouraged many researchers to explore ways to execute protocols in parallel, or to design new "lightweight" protocols. Other sources of problems were thought to be poor protocol implementations, high overheads associated with operating system functions, and a generally poor interface between applications and the network services.Clark et al. [l] suggested that even heavyweight protocols, such as the widely used Transmission Control Protocol/Internet Protocol (TCP/IP) combination, could be extremely efficient if implemented sensibly. More recently, Jacobson has shown that most T C P i I P packets can be processed by fewer than 100 instructions [2]. It is now widely believed that while a poor implementation will impede performance, protocols such as TCP are not inherent limiting factors.One reason many implementations fail to achieve high throughput is that they access user data several times between the instant the data are generated and the instant the data are transmitted on the network. In the rest of this article,we analyze this behavior in a widely used implementation of TCP and consider three proposals for improving its performance. We describe o u r experimental implementation of one of these proposals, which uses novel hardware together with a revised implementation of the protocol. T o conclude, we present measurements of the system's performance. ries in Bristol, U.K.The Bottleneck: Copying Data e believe that the speed of protocol imple-w mentations in current workstations is limited not by their calculation rate, but by how quickly they can move data. This section first reviews the design of a popular protocol implementation and then examines its behavior with reference t o workstation performance.Reception -T h e driver copies an incoming packet into a kernel buffer, then starts it moving through the protocol receive functions. Most of these only look at the headers. However, tcp-input reads all the data in the packet to calculate a checksum to compare with the one in the header. It places valid data in a queue for the appropriate socket, again using pointers rather than copying. 36IEEE Network July 1993 OXY0-8044/92/$03.~~0 0 1993 IEEE
Highly customised variable-data documents make automatic layout of the resulting publication hard. Architectures for defining and processing such documents can benefit if the repertoire of layout methods available can be extended smoothly and easily to accommodate new styles of customisation. The Document Description Framework incorporates a model for declarative document layout and processing where documents are treated as functional programs. A canonical XML tree contains nodes describing layout instructions which will modify and combine their children component parts to build sections of the final presentation. Leaf components such as images, vector graphic fragments and text blocks are 'rendered' to make consistent graphical atoms. These parts are then processed by layout agents, described and parameterised by their parent nodes, which can range from simple layouts like translations, flows, encapsulations and tables through to highly complex arrangements such as constraint-solution or pagination. The result then becomes a 'molecule' for processing at a higher level of the layout tree. A variable and reference mechanism is included for resolving rendering interdependency and supporting component reuse. Addition of new layout types involves definition of a new combinator node and attachment of a suitable agent.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.