Pipelining parallel image compositing and delivery for efficient remote visualization

Scientific datasets of large volumes generated by next-generation computational sciences need to be transferred and processed for remote visualization and distributed collaboration among a geographically dispersed team of scientists. Parallel visualization using high-performance computing facilities is a typical approach to processing such increasingly large datasets. We propose an optimized image compositing scheme with linear pipeline and adaptive transport to support efficient image delivery to a remote client. The proposed scheme arranges an arbitrary number of parallel processors within a cluster in a linear order and divides the image into a carefully selected number of segments, which flow through the linear in-cluster pipeline and wide-area networks to the remote client consecutively. We analytically determine the segment size that minimizes the final image display time and derive the conditions where the proposed image compositing and delivery scheme outperforms the traditional schemes including the binary swap algorithm. In order to match the transport throughput for image delivery over wide-area networks to the pipelining rate for image compositing within the cluster, we design a class of transport protocols using stochastic approximation methods that are able to stabilize the data flow at a target rate. The experimental results from remote visualization of large-scale scientific datasets justify the correctness of our theoretical analysis and illustrate the superior performances of the proposed method.     

Long-haul secure data transfer using hardware-assisted GridFTP

Extreme-scale scientific collaborations require high-performance wide-area end-to-end data transports to enable fast and secure transfer of high data volumes among collaborating institutions. GridFTP is the de facto protocol for large-scale data transfer in science environments. Existing predominant network transport protocols such as TCP have serious limitations that consume significant CPU power and prevent GridFTP from achieving high throughput on long-haul networks with high latency and potential packet loss, reordering and jitter. On the other hand, protocols such as UDT that address some of the TCP shortcomings demand high computing resources on data transfer nodes. These limitations have caused underutilization of existing high-bandwidth links in scientific and collaborative grids. To address this situation, we have enhanced Globus GridFTP, the most widely used GridFTP implementation, by developing transport offload engines such as UDT and iWARP on SmartNIC, a programmable 10GbE network interface card (NIC). Our results show significant reduction in server utilization and full line-rate sustained bandwidth in high-latency networks, as measured for up to 100 ms of network latency. In our work, we also offload OpenSSL on SmartNIC to reduce host utilization for secure file transfers. The offload engine can provide line-rate data channel encryption/decryption on top of UDT offload without consuming additional host CPU resources. Lower CPU utilization leads to increased server capacity, which allows data transfer nodes to support higher network and data-processing rates. Alternatively, smaller or fewer DTNs can be used for a particular data rate requirement.     

Dynamic window flow control on a high-speed wide-area data network

This paper considers sliding-window flow control for data traffic on a high-speed, wide-area, virtual-circuit backbone network. The objectives are to allocate network bandwidth fairly, to utilize bandwidth efficiently, and to prevent buffer overflow within the network, but without requiring gross overengineering of memory in the switching nodes. The network uses round robin queueing disciplines at routers and switches. A signaling protocol is described for in-call adjustment of windows and per-virtual-circuit buffer allocations as a function of network congestion, while maintaining consistent allocations along each virtual circuit. Simulations indicate that the protocol works as expected between the edges of the wide-area network. The wide-area backbone may interconnect local-area networks on which the hosts use transport-level protocols such as TCP. The backbone network protocol works satisfactorily with the host-to-host transport provided that the hosts use a retransmission strategy such as Jacobson-Karels to deal with possible packet losses at the edge of the backbone network.     

Phoebus: A system for high throughput data movement

Phoebus is an infrastructure for improving end-to-end throughput in high-bandwidth, long-distance networks by using a “session layer” protocol and “gateways” in the network. Phoebus has the ability to dynamically allocate network resources and to use segment-specific transport protocols between gateways, as well as to apply other performance-improving techniques on behalf of the user. We have developed interfaces to Phoebus to allow its use in various real applications and data movement services. This paper extends our earlier work with tests of Phoebus-enabled applications on both real-world networks as well as over configurable network testbeds that allow us to modify latency and loss rates. We demonstrate that Phoebus improves the performance of bulk data transfer in a variety of network configurations and conditions.     

Parallel transport subsystem implementation for high-performance communication

Requirements of emerging applications together with rapid changes in networking technology towards gigabit speeds require new adequate transport systems. Integrated designs of transport services, protocol architecture and implementation platforms are required by forthcoming applications in high-speed network environments. The transport subsystem PATROCLOS (parallel transport subsystem for cel/ based high-speed networks) is designed with special emphasis on a high degree of inherent parallelism to allow efficient implementations on multiprocessor architectures combined with specialized hardware for very time critical functions. The paper presents the new parallel protocol architecture of PATROCLOS, an appropriate implementation architecture based on transputer networks, and performance evaluation results, which indicate high throughput values.     

SCTP: new transport protocol for TCP/IP

For the past 20 years (1980-2000), applications and end users of the TCP/IP suite have employed one of two protocols: the transmission control protocol or the user datagram protocol. Yet some applications already require greater functionality than what either TCP or UDP has to offer, and future applications might require even more. To extend transport layer functionality, the Internet Engineering Task Force approved the stream control transmission protocol (SCTP) as a proposed standard in October 2000. SUP was spawned from an effort started in the IETF Signaling Transport (Sigtrans) working group to develop a specialized transport protocol for call control signaling in voice-over-IP (VoIP) networks. Recognizing that other applications could use some of the new protocol's capabilities, the IETF now embraces SCTP as a general-purpose transport layer protocol, joining TCP and UDP above the IP layer. Like TCP, STCP offers a point-to-point, connection-oriented, reliable delivery transport service for applications communicating over an IP network     

Socket Buffer Auto-Sizing for High-Performance Data Transfers

It is often claimed that TCP is not a suitable transport protocol for data intensive Grid applications in high-performance networks. We argue that this is not necessarily the case. Without changing the TCP protocol, congestion control, or implementation, we show that an appropriately tuned TCP bulk transfer can saturate the available bandwidth of a network path. The proposed technique, called SOBAS, is based on automatic socket buffer sizing at the application layer. In non-congested paths, SOBAS limits the socket buffer size based on direct measurements of the received throughput and of the corresponding round-trip time. The key idea is that the send window should be limited, after the transfer has saturated the available bandwidth in the path, so that the transfer does not cause buffer overflows (self-induced losses). A difference with other socket buffer sizing schemes is that SOBAS does not require prior knowledge of the path characteristics, and it can be performed while the transfer is in progress. Experimental results in several high bandwidth-delay product paths show that SOBAS provides consistently a significant throughput increase (20% to 80%) compared to TCP transfers that use the maximum possible socket buffer size. We expect that SOBAS will be mostly useful for applications such as GridFTP in non-congested wide-area networks.     

A systematic approach to host interface design for high-speednetworks

Optical fiber has made it possible to build networks with link speeds of over a gigabit per second; however, these networks are pushing end-systems to their limits. For high-speed networks (100 Mbits per second and up), network throughput is typically limited by software overhead on the sending and receiving hosts. Minimizing this overhead improves application-level latency and throughput and reduces the number of cycles that applications lose to communication overhead. Several factors influence communication overhead: communication protocols, the application programming interface (API). and the network interface hardware architecture. The author describes how these factors influence communication performance and under what conditions hardware support on the network adapter can reduce overhead. He first describes the organization of a typical network interface and discusses performance considerations for interfaces to high-speed networks. He then discusses software optimizations that apply to simple network adapters and show how more powerful adapters can improve performance on high-speed networks     

Dependable user-level socket over dual networks

Message duplication over multiple links can enhance the communication reliability and availability among distributed processes running on clusters or networked workstations. In addition, message striping over multiple links can enhance communication throughput and transfer times. This paper introduces optimized techniques to provide a dependable user-level socket that enhances the reliability and performance of message deliveries over dual network interfaces. The proposed techniques are designed to work on top of reliable transport protocols. In addition, a technique for efficient duplication of messages on dual networks to enhance both communication reliability and performance on top of reliable transport protocol is proposed. A prototype socket based on this model is developed and evaluated. The experimental evaluations of the proposed model indicate good communication reliability and performance gains over conventional methods.     

Distributed Throughput Optimization for Large-Scale Scientific Workflows Under Fault-Tolerance Constraint

With the advent of next-generation scientific applications, the workflow approach that integrates various computing and networking technologies has provided a viable solution to managing and optimizing large-scale distributed data transfer, processing, and analysis. This paper investigates a problem of mapping distributed scientific workflows for maximum throughput in faulty networks where nodes and links are subject to probabilistic failures. We formulate this problem as a bi-objective optimization problem to maximize both throughput and reliability. By adapting and modifying a centralized fault-free workflow mapping scheme, we propose a new mapping algorithm to achieve high throughput for smooth data flow in a distributed manner while satisfying a pre-specified bound of the overall failure rate for a guaranteed level of reliability. The performance superiority of the proposed solution is illustrated by both extensive simulation-based comparisons with existing algorithms and experimental results from a real-life scientific workflow deployed in wide-area networks.     

High-performance message striping over reliable transport protocols

This paper introduces a high-performance middleware-level message striping approach to increase communication bandwidth for data transfer in heterogeneous clusters equipped with multiple networks. In this scheme, concurrency is used for the striping process. The proposed striping approach is designed to work at the middleware-level, between the distributed applications and the reliable transport protocols such as TCP. The middleware-level striping approach provides flexible, scalable, and hardware-, network-, and operating systems-independent communication bandwidth solution. In addition, techniques to enhance the performance of this approach over multiple networks are introduced. The proposed techniques, which minimize synchronization contention and eliminate the striping sequence header, rely on the features of a reliable transport protocol such as TCP to reduce some of the concurrent striping overhead. The techniques have been implemented and evaluated on a real cluster with multiple networks and the results show significant performance gains for data transfer over existing approaches.     

An on-demand routing protocol for improving channel use efficiency in multichannel ad hoc networks

In wireless ad hoc networks, end-to-end delivery over network is a critical concern for routing protocols. The capacity of routing protocols is constrained by the intra-flow interference introduced by adjacent nodes on the same path, and inter-flow interference generated by nodes from neighboring paths. In this paper, we develop an on-demand routing protocol M-AODV-R that solves the channel assignment, reuse and routing problem jointly. The proposed channel reuse scheme and channel assignment scheme can enhance channel reuse rate. This cross-layer design approach can significantly improve the performance of multichannel ad hoc networks over existing routing protocols. Simulation results show that the proposed routing M-AODV-R can effectively increase throughput and reduce delay, as compared to AODV protocol.     

Optimizing end-to-end performance of data-intensive computing pipelines in heterogeneous network environments

Supporting high-performance data-intensive computing pipelines in wide-area networks is crucial for enabling large-scale distributed scientific applications that require minimizing end-to-end delay for single-input applications or maximizing frame rate for streaming applications. We formulate and categorize the data-intensive computing pipeline mapping problems into six classes with two optimization objectives, i.e. minimum end-to-end delay and maximum frame rate, and three network constraints, i.e. no, contiguous, and arbitrary node reuse. We design a dynamic programming-based optimal solution to the problem of minimum end-to-end delay with arbitrary node reuse and prove the NP-completeness of the rest five problems, for each of which, a heuristic algorithm based on a similar optimization procedure is proposed. These heuristics are implemented and tested on a large set of simulated pipelines and networks of various scales and their performance superiority is illustrated by extensive simulation results in comparison with existing methods.     

通信协议的并行机制研究

通信技术的迅速发展，通信网络所提供的带宽达到几百Ｍｂｐｓ～Ｇｂｐｓ数量级。一些应用，象多媒体通信，就能利用网络的高吞吐量。但是，目前的通信协议并不能适应高速传输的环境。运输层和其它较高层协议的处理，已成为通信系统的性能瓶颈。本文首先回顾近十年来通信环境所发生的一些变化，并分析其对运输层协议的影响；第二部分分析标准协议并行实现时的一些关键技术；最后介绍了一个运输层系统模型，该模型采用并行机制，向上提供高性能的运输层服务。     

Multi-path continuous media streaming: what are the benefits?

Quality of service (QoS) in delivery of continuous media (CM) over the Internet is still relatively poor and inconsistent. Although many such applications can tolerate some degree of missing information, significant losses degrade an application’s QoS. In this paper, we investigate the potential benefits of mitigating this problem through the exploitation of multiple paths existing in the network between a set of senders and a receiver of CM. Our focus in this work is on providing a fundamental understanding of the benefits of using multiple paths to deliver CM over best-effort wide-area networks. Specifically, we consider pre-recorded CM applications and use the following metrics in evaluating the performance of multi-path streaming as compared to single-path streaming: (a) data loss rate, (b) conditional error burst length distribution, and (c) lag1-autocorrelation. The results of this work can be used in guiding the design of multi-path CM systems streaming data over best-effort wide-area networks.     

Nonblocking WDM multicast switching networks

With ever increasing demands on bandwidth from emerging bandwidth-intensive applications, such as video conferencing, E-commerce, and video-on-demand services, there has been an acute need for very high bandwidth transport network facilities. Optical networks are a promising candidate for this type of applications. At the same time, many bandwidth-intensive applications require multicast services for efficiency purposes. Multicast has been extensively studied in the parallel processing and electronic networking community and has started to receive attention in the optical network community recently. In particular, as WDM (wavelength division multiplexing) networks emerge, supporting WDM multicast becomes increasingly attractive. In this paper, we consider efficient designs of multicast-capable WDM switching networks, which are significantly different and, hence, require nontrivial extensions from their electronic counterparts. We first discuss various multicast models in WDM networks and analyze the nonblocking multicast capacity and network cost under these models. We then propose two methods to construct nonblocking multistage WDM networks to reduce the network cost     

Improving End System Performance for Multimedia Applications over High Bandwidth Networks

Digital video services, scientific visualization and other multimedia applications require delivery of high network throughput to end user applications. In this paper we identify bottlenecks in the data path between high-speed networks and applications. Using performance of multimedia applications as a metric, the effectiveness of solutions to reduce network, operating system, and user bottlenecks is explored experimentally.     

群组通信模型及运输协议映射*

新型网络应用要求通信协议提供多点投递和相应群组管理功能,同时,高速传输服务和新型网络层协议也开始具有数据多点传输和简单群组控制能力,跨越两者的运输层协议,从而又重新成为学术研究和标准化的热点和趋势.文章主要描述建立群组通信抽象模型的过程和结论以及参照新型运输协议XTP(express transport protocol)和计算机会议应用的模型映射和评价.     

QoS based multicast routing algorithms for real time applications

In recent years, there has been a lot of interest in providing real-time multimedia services like digital audio and video over packet-switched networks such as Internet and ATM. These services require certain quality of service (QoS) from the network. The routing algorithm should take QoS factor for an application into account while selecting the most suitable route for the application. In this paper, we introduce a new routing metric and use it with two different heuristics to compute the multicast tree for guaranteed QoS applications that need firm end-to-end delay bound. We then compare the performance of our algorithms with the other proposed QoS-based routing algorithms. Simulations were run over a number of random networks to measure the performance of different algorithms. We studied routing algorithms along with resource reservation and admission control to measure the call throughput over a number of random networks. Simulation results show that our algorithms give a much better performance in terms of call throughput over other proposed schemes.     

Analysis of a Hybrid (demand assignment TDMA) protocol for video teleconferencing-voice data optical networks

An optical star connected backbone network has recently been designed [1] and is currently being built. Though the network is local, it is mainly intended for internetworking applications. A large number of local networks such as Ethernet or other integrated voice data networks can be interconnected via the local optical network. Also, because of the enormous bandwidth available, point to point and conference video users can be easily accomodated (simultaneously operating with the other applications). To be able to accommodate such huge services, the central switch and the 16 data concentrators of the proposed network have a great deal of software protocols to execute. The heart of these procedures is a hybrid (demand assignment — TDMA) protocol. The object of this paper is to give a concise picture of the network, the protocols involved and more importantly, to evaluate the throughput efficiency and delay characteristics.