Data Transfer Performance Research Articles

On performance bounds for balanced fairness

While Erlang’s formula has helped engineers to dimension telephone networks for over 80 years, such a three-way “performance–demand–capacity” relationship is still lacking for data networks. It may be argued that the enduring success of Erlang’s formula is essentially due to its simplicity: the call blocking rate does not depend on the distribution of call duration but on overall demand only. In this paper, we consider data networks and characterize those capacity allocations which have the same insensitivity property, in the sense that performance of data transfers does not depend on precise traffic characteristics such as the distribution of data volume but on overall demand only. We introduce the notion of “balanced fairness” and prove some key properties satisfied by this insensitive allocation. It is shown notably that the performance of balanced fairness is always better than that obtained if flows are transmitted in a “store and forward” fashion, allowing simple formula applying to the latter to be used as a conservative evaluation for network design and provisioning purposes.

Evaluation of Applicability of Highly reliable Duplex Systems that employ Fault-tolerant Microprocessors.

We evaluated the functions and the performance of a highly reliable system which is the tightly-coupled duplex system made up of two microprocessors, the GMICRO/100 and the G100FTS (GMICRO/100 for Fault-Tolerant System) which is a fault tolerant microprocessor.During normal operation mode, normal processing of data transfer and further more, processing for recovery from a transient fault and a permanent fault as well as maintenance of processing continuity have been confirmed.It has been demonstrated that this system enables us to obtain high reliability. On the other hand, it was revealed that the average performance of numerical calculation and data transfer of this system was 69 to 75 % that of a simplex system. From this result, it was found that when the duplex system is to be applied to an application system that requires real time operations, it is necessary to recognize in advance the relationship between the tolerable maximum response time and the feasible fault-tolerant function of the application system.

Quality of Service Networking for High Performance Grid Applications

This paper reports on different efforts to provide quality of service (QoS) Networking to Grid applications done in the context of the MB-NG, GRS and DataTAG EU projects. These are leading edge network research projects involving more that 50 researchers in the UK, Europe and North America, concerned with the development and testing of protocols and standards for the next generation of high speed networks. We have implemented and tested the Differentiated Services Architecture (DiffServ) in a multi-domain, 2.5 Gbits/s network (the first such deployment) defining appropriate Service Level Agreements (SLAs) to be used between administrative domains to guarantee end-to-end Quality of Service. We characterised several hardware implementations of DiffServ and concluded on their appropriateness for several network scenarios. Since current and future Grid applications will have to use modified mechanisms of congestion control we have evaluated old and new TCP implementations over a Differentiated Services Networks. These quality of service tests have also included innovative MPLS (Multi-Protocol Label Switching) experiments to establish guaranteed bandwidth connections to Grid applications in a fast and efficient way. We have also developed a software based bandwidth broker architecture for Grids based on IETF standards which allows applications to transparently request dynamic and advanced reservations and implemented it in a real experimental network. We finally report on experiences delivering Quality of Service networking to high performance applications like Particle Physics data transfer and High Performance Computation. This includes quantitative results on the performance improvements that QoS brought to real data transfers in the context of High Performance Computing.

Early experiences with the GridFTP protocol using the GRB-GSIFTP library

The ISUFI/High Performance Computing Centre is actively experimenting with a grid infrastructure connecting sites located both in Europe and in the USA. This paper reports on our feasibility study related to the use of the GridFTP protocol for high performance data transfers. We also describe a client library we are developing to provide additional services beyond those provided by the Globus ftp client library and to ease code development.

Head-media interface instability under hostile operating conditions

In the office environment, computers reliably transfer data to and from the hard disk drive and data transfer errors are extremely rare. However, when even ruggedized computers are operated in hostile conditions, valuable data can be lost as a result of read-write error due to either shock or vibration, seriously impairing drive performance. For this work, drives have been vibrated over large frequency ranges in order to characterize their data-transfer performance. In order to observe the cause of the problems, a drive has been modified with integral sensors to identify internal system failures and the results of these investigations are presented. These sensors have enabled the first unobtrusive tests of the drives' mechanical frequency response while the drive is in operation with sensors that can be easily and inexpensively embedded within a hard disk drive for use in a generic, smart, vibration resilient drive.

Critical path analysis of TCP transactions

Improving the performance of data transfers in the Internet (such as Web transfers) requires a detailed understanding of when and how delays are introduced. Unfortunately, the complexity of data transfers like those using HTTP is great enough that identifying the precise causes of delays is difficult. In this paper, we describe a method for pinpointing where delays are introduced into applications like HTTP by using critical path analysis. By constructing and profiling the critical path, it is possible to determine what fraction of total transfer latency is due to packet propagation, network variation (e.g., queueing at routers or route fluctuation), packet losses, and delays at the server and at the client. We have implemented our technique in a tool called tcpeval that automates critical path analysis for Web transactions. We show that our analysis method is robust enough to analyze traces taken for two different TCP implementations (Linux and FreeBSD). To demonstrate the utility of our approach, we present the results of critical path analysis for a set of Web transactions taken over 14 days under a variety of server and network conditions. The results show that critical path analysis can shed considerable light on the causes of delays in Web transfers, and can expose subtleties in the behavior of the entire end-to-end system.

Critical path analysis of TCP transactions

Improving the performance of data transfers in the Internet (such as Web transfers) requires a detailed understanding of when and how delays are introduced. Unfortunately, the complexity of data transfers like those using HTTP is great enough that identifying the precise causes of delays is difficult. In this paper we describe a method for pinpointing where delays are introduced into applications like HTTP by using critical path analysis. By constructing and profiling the critical path, it is possible to determine what fraction of total transfer latency is due to packet propagation, network variation ( e.g. , queuing at touters or route fluctuation), packet losses, and delays at the server and at the client. We have implemented our technique in a tool called tcpeval that automates critical path analysis for Web transactions. We show that our analysis method is robust enough to analyze traces taken for two different TCP implementations (Linux and FreeBSD). To demonstrate the utility of our approach, we present the results of critical path analysis for a set of Web transactions taken over 14 days under a variety of server and network conditions. The results show that critical path analysis can shed considerable light on the causes of delays in Web transfers, and can expose subtleties in the behavior of the entire end-to-end system.

Model of Head-Positioning Error Due to Rotational Vibration of Hard Disk Drives

An analytical model of head-positioning error due to rotational vibration of a hard disk drive is proposed. The model takes into account the rotational vibration of the base plate caused by the reaction force of the head-positioning actuator, the relationship between the rotational vibration and head-track offset, and the sensitivity function of track-following feedback control. Error calculated by the model agrees well with measured error. It is thus concluded that this model can predict the data transfer performance of a disk drive in read mode.

Critical path analysis of TCP transactions

Improving the performance of data transfers in the Internet (such as Web transfers) requires a detailed understanding of when and how delays are introduced. Unfortunately, the complexity of data transfers like those using HTTP is great enough that identifying the precise causes of delays is difficult. In this paper we describe a method for pinpointing where delays are introduced into applications like HTTP by using critical path analysis. By constructing and profiling the critical path, it is possible to determine what fraction of total transfer latency is due to packet propagation, network variation ( e.g., queuing at routers or route fluctuation), packet losses, and delays at the server and at the client. We have implemented our technique in a tool called tcpeval that automates critical path analysis for Web transactions. We show that our analysis method is robust enough to analyze traces taken for two different TCP implementations (Linux and FreeBSD). To demonstrate the utility of our approach, we present the results of critical path analysis for a set of Web transactions taken over 14 days under a variety of server and network conditions. The results show that critical path analysis can shed considerable light on the causes of delays in Web transfers, and can expose subtleties in the behavior of the entire end-to-end system.

Model of Head-Positioning Error due to Rotational Vibration of Hard Disk Drives.

An analysis method of a head-positioning error due to rotational vibration of magnetic disk drive is proposed. The model takes into account the rotational vibration of the baseplate caused by the reaction force of the head-positioning actuator, the relationship between the rotational vibration and head-track offset, and the sensitivity function of the track-following feedback control. Simulation resilts obtained using the model coincide with experimental data. This model enabled us to predict the data transfer performance.

Performance of high‐speed data transfer using an fec method in a multimedia network environment

AbstractIn the ATM system, all multimedia information, such as voice, picture and data, is divided into packets with the fixed length of 53 bytes, called cells, and the cells are transferred. Intensive studies have been made on the performance of the system, where such multimedia information is multiplexed in the network switch. In the ATM network, the cell loss is not compensated, and it is necessary that the end‐user should recover from the errors in order to realize highly reliable data transfer among computers. As one such approach, the error‐correcting code (FEC) has been considered interesting.From such a viewpoint, this paper evaluates analytically the data transfer performance when FEC is used. When FEC is used, the redundant cell produces an overhead, which may enhance congestion when the traffic in the network is high and deteriorate the performance. Also, from this viewpoint, this paper considers the priority control, where the redundant cells are discarded positively when the network is congested. The effectiveness of the approach is examined, as in the data transfer performance for the case where the video and the data exist together in the traffic, and it is demonstrated that the control parameters in the priority control are especially important in that case.

The Leading Edge in High-Speed Data Communications Services: The InterSpan® Asynchronous Transfer Mode Service

The emergence of new high-speed client/server and multimedia applications demands a new generation of computer communications services. The InterSpan® Asynchronous Transfer Mode (ATM) Service, first introduced in May 1994, is a highly reliable, low-cost, shared transport service for these and other applications. Using the ATM interface protocol, the service provides multiple grades of service over a wide range of transmission speeds (currently up to 45 Mb/s), ensuring a smooth evolution to the future. The service has three main features that differentiate it from the competition: the ReliaBURST <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">SM</sup> traffic management method, which improves data transfer performance; an industry-leading Customer Network Management System, which lets customers “see” their portion of the shared network; and a frame relay-to-ATM protocol conversion service, which makes the evolution to ATM easier by protecting the investment of customers currently using the InterSpan Frame Relay Service. The service development plan has a rich set of additional features, including higher speeds, switched virtual circuits for high-capacity bandwidth on demand, and globalization.

Development of Monte Carlo Machine for Particle Transport Problem.

Monte Carlo machine, Monte-4 has been developed to realize high performance computing of Monte Carlo codes for particle transport. The calculation for particle tracking in a complex geometry requires (1) classification of particles by the region types using multi-way conditional branches, and (2) determination whether intersections of particle paths with surfaces of the regions are on the boundaries of the regions or not, using nests of conditional branches. How-ever, these procedures require scalar operations or unusual vector operations. Thus the speedup ratios have been low, i.e. nearly two times, in vector processing of Monte Carlo codes for particle transport on conventional vector processors. The Monte Carlo machine Monte-4 has been equipped with the special hardware called Monte Carlo pipelines to process these procedures with high performance. Additionally Monte-4 has been equipped with enhanced load/store pipelines to realize fast transfer of indirectly addressed data for the purpose of resolving imbalances between the performance of data transfers and arithmetic operations in vector processing of Monte Carlo codes on conventional vector processors. Finally, Monte-4 has a parallel processing capability with four processors to multiply the performance of vector processing. We have evaluated the effective performance of Monte-4 using production-level Monte Carlo codes such as vectorized KENO-IV and MCNP. In the performance evaluation, nearly ten times speedup ratios have been obtained, compared with scalar processing of the original codes.

Distributed scheduling algorithms to improve the performance of parallel data transfers

The cost of data transfers, and in particular of I/O operations, is a growing problem in parallel computing. A promising approach to alleviating this bottleneck is to schedule parallel I/O operations explicitly. We develop a class of decentralized algorithms for scheduling parallel I/O operations, where the objective is to reduce the time required to complete a given set of transfers. These algorithms, based on edge-coloring and matching of bipartite graphs, rely upon simple heuristics to obtain shorter schedules. We present simulation results indicating that the best of our algorithms can produce schedules whose length is within 2--20% of the optimal schedule, a substantial improvement on previous decentralized algorithms. We discuss theoretical and experimental work in progress and possible extensions.

A Database for Environmental Research Programmes

Good-quality, long-term data sets are invaluable in researching environmental change. A relational database management system and a geographical information system have been loosely integrated to provide researchers, working in an environmental research program, with access to such potentially important data sets. Due to varying data quality, missing information and the diversity of the data, metadata plays a key role in the system design. The environmental research database incorporates two inter-related databases, containing meta-information and data for analyses, respectively. A menu-driven user interface enables geographically remote researchers, connected to an academic-wide area network, to access these databases. Due to the complexity of the geographic information system and the need to perform data transfers, users must submit requests, via the menu system, to take advantage of such "off-line" facilities. Proposed improvements in information systems technology will permit the geographical information system and the relational database management system to be integrated further, as well as reducing data management effort.

Interactive numberical simulation and grid adaption for hypersonic flow

The numerical simulation of hypersonic flows requires the use of supercomputers and graphic workstations in an efficient manner. The present contribution reports on an implementation of a finite element Navier-Stokes method with grid adaption, by which user interaction is used to optimize efficiency of high performance computing in a supercomputer/workstation network. Scaling laws are derived to predict the feasibility of this technique on other computers and networks with different CPU and data transfer performance. The High Performance Computer Network of Lower Saxony is used as an example for the application of these scaling laws. We find that storage and CPU performance of this example as well as at many other supercomputer sites is sufficient for fine adaptive grids to be used, but data transfer performance must be increased. High speed data links between supercomputers and workstations can increase efficiency of high performance computing at relatively low costs.

Optimizing bulk data transfer performance: a packet train model

Optimizing bulk data transfer performance: a packet train model

Connection-oriented versus connectionless protocols: A performance study

The impact of various implementations of an end-to-end transport service on the performance of high-volume data transfers in a local-area network is investigated. The impact of using a connection-oriented logical-link control (LLC) protocol below a connection-oriented transport protocol is of particular interest. The performance measure is the throughput which can be achieved between two stations either from memory to memory or from disk to disk. The throughput was determined by simulation for various combinations of differing LLC protocols, disk models and processor speeds. The parameters for the simulation model were determined partially by measurements of an implemented file server. The results show that the performance penalty of a connection-oriented LLC protocol can be substantial, but can be significantly lowered when an acknowledgment accumulation strategy is applied.