Showing papers on "Scalability published in 1996"

SPRINT: A Scalable Parallel Classifier for Data Mining

Abstract: Classification is an important data mining problem. Although classification is a wellstudied problem, most of the current classification algorithms require that all or a portion of the the entire dataset remain permanently in memory. This limits their suitability for mining over large databases. We present a new decision-tree-based classification algorithm, called SPRINT that removes all of the memory restrictions, and is fast and scalable. The algorithm has also been designed to be easily parallelized, allowing many processors to work together to build a single consistent model. This parallelization, also presented here, exhibits excellent scalability as well. The combination of these characteristics makes the proposed algorithm an ideal tool for data mining.

A fast distributed algorithm for mining association rules

Abstract: With the existence of many large transaction databases, the huge amounts of data, the high scalability of distributed systems, and the easy partitioning and distribution of a centralized database, it is important to investigate efficient methods for distributed mining of association rules. The study discloses some interesting relationships between locally large and globally large item sets and proposes an interesting distributed association rule mining algorithm, FDM (fast distributed mining of association rules), which generates a small number of candidate sets and substantially reduces the number of messages to be passed at mining association rules. A performance study shows that FDM has a superior performance over the direct application of a typical sequential algorithm. Further performance enhancement leads to a few variations of the algorithm.

Distributed shared memory: concepts and systems

Abstract: In surveying current approaches to distributed shared memory computing, the authors find that the reduced cost of parallel software development will help make the DSM paradigm a viable solution to large scale, high performance computing.

Global arrays: a nonuniform memory access programming model for high-performance computers

Abstract: Portability, efficiency, and ease of coding are all important considerations in choosing the programming model for a scalable parallel application. The message-passing programming model is widely used because of its portability, yet some applications are too complex to code in it while also trying to maintain a balanced computation load and avoid redundant computations. The shared-memory programming model simplifies coding, but it is not portable and often provides little control over interprocessor data transfer costs. This paper describes an approach, called Global Arrays (GAs), that combines the better features of both other models, leading to both simple coding and efficient execution. The key concept of GAs is that they provide a portable interface through which each process in a MIMD parallel program can asynchronously access logical blocks of physically distributed matrices, with no need for explicit cooperation by other processes. We have implemented the GA library on a variety of computer systems, including the Intel Delta and Paragon, the IBM SP-1 and SP-2 (all message passers), the Kendall Square Research KSR-1/2 and the Convex SPP-1200 (nonuniform access shared-memory machines), the CRAY T3D (a globally addressable distributed-memory computer), and networks of UNIX workstations. We discuss the design and implementation of these libraries, report their performance, illustrate the use of GAs in the context of computational chemistry applications, and describe the use of a GA performance visualization tool.

A scalable and highly available web server

Abstract: We describe a prototype scalable and highly available web server, built on an IBM SP-2 system, and analyze its scalability. The system architecture consists of a set of logical front-end or network nodes and a set of back-end or data nodes connected by a switch, and a load balancing component. A combination of TCP routing and Domain Name Server (DNS) techniques are used to balance the load across the Front-end nodes that run the Web (httpd) daemons. The scalability achieved is quantified and compared with that of the known DNS technique. The load on the back-end nodes is balanced by striping the data objects across the back-end nodes and disks. High availability is provided by detecting node or daemon failures and reconfiguring the system appropriately. The scalable and highly available web server is combined with parallel databases, and other back-end servers, to provide integrated scalable and highly available solutions.

A predicate-based caching scheme for client-server database architectures

Abstract: We propose a new client-side data-caching scheme for relational databases with a central server and multiple clients. Data are loaded into each client cache based on queries executed on the central database at the server. These queries are used to form predicates that describe the cache contents. A subsequent query at the client may be satisfied in its local cache if we can determine that the query result is entirely contained in the cache. This issue is called cache completeness. A separate issue, cache currency, deals with the effect on client caches of updates committed at the central database. We examine the various performance tradeoffs and optimization issues involved in addressing the questions of cache currency and completeness using predicate descriptions and suggest solutions that promote good dynamic behavior. Lower query-response times, reduced message traffic, higher server throughput, and better scalability are some of the expected benefits of our approach over commonly used relational server-side and object ID-based or page-based client-side caching.

MONET: multiwavelength optical networking

Abstract: This paper presents an overview of the multiwavelength optical networking (MONET) program and summarizes its vision. The program objective is to advance, demonstrate, and integrate network architecture and economics, advanced multiwavelength technology, and network management and control to achieve high capacity, reconfigurable, high performance, reliable multiwavelength optical networks, with scalability to national scale, for both commercial and specialized government applications. The paper describes the major research thrusts of the program including network elements, networking demonstration plans, network control and management, and architecture and economics.

Bridge/router architecture for high performance scalable networking

Abstract: A high performance scalable networking bridge/router system is based on a backbone communication medium and message passing process which interconnects a plurality of input/output modules. The input/output modules vary in complexity from a simple network interface device having no switching or routing resources on board, to a fully functional bridge/router system. Also, in between these two extremes input/output modules which support distributed protocol processing are supported. A central internetworking engine includes a shared memory resource coupled to the backbone. The architecture includes a physical layer communication system for transferring control messages and data packets across the backbone, a logical layer interprocessor messaging system which operates over the physical layer across the backbone for communication between intelligent input/out modules, and between such modules in the central internetworking engine, and distributed protocol modules which are supported on intelligent input/output modules, and communicate using the logical interprocessor messaging system with the central internetworking resources.

The Nimrod Routing Architecture

Abstract: We present a scalable internetwork routing architecture, called Nimrod. The Nimrod architecture is designed to accommodate a dynamic internetwork of arbitrary size with heterogeneous service requirements and restrictions and to admit incremental deployment throughout an internetwork. The key to Nimrod's scalability is its ability to represent and manipulate routing-related information at multiple levels of abstraction.

CoBase: a scalable and extensible cooperative information system

Abstract: A new generation of information systems that integrates knowledge base technology with database systems is presented for providing cooperative (approximate, conceptual, and associative) query answering. Based on the database schema and application characteristics, data are organized into Type Abstraction Hierarchies (TAHs). The higher levels of the hierarchy provide a more abstract data representation than the lower levels. Generalization (moving up in the hierarchy), specialization (moving down the hierarchy), and association (moving between hierarchies) are the three key operations in deriving cooperative query answers for the user. Based on the context, the TAHs can be constructed automatically from databases. An intelligent dictionary/directory in the system lists the location and characteristics (e.g., context and user type) of the TAHs. CoBase also has a relaxation manager to provide control for query relaxations. In addition, an explanation system is included to describe the relaxation and association processes and to provide the quality of the relaxed answers. CoBase uses a mediator architecture to provide scalability and extensibility. Each cooperative module, such as relaxation, association, explanation, and TAH management, is implemented as a mediator. Further, an intelligent directory mediator is provided to direct mediator requests to the appropriate service mediators. Mediators communicate with each other via KQML. The GUI includes a map server which allows users to specify queries graphically and incrementally on the map, greatly improving querying capabilities. CoBase has been demonstrated to answer imprecise queries for transportation and logistic planning applications. Currently, we are applying the CoBase methodology to match medical image (X-ray, MRI) features and approximate matching of emitter signals in electronic warfare applications.

Parallel and Distributed Simulation

Abstract: Parallel and distributed simulation (PDS) over the one and a half decade of its existence has turned out to be more foundational than merely solving the causality preservation di lemma in partially ordered event structures as they occur in parallel and distributed simulation executions. Today’s availability of parallel and distributed computing and communication technology ha.s given a new relevance to the field that could not, ha.ue been foreseen in its early days. The potentia,l improvement of elapsed time for large simula.tion experiments via the involvement of a set of individua.1 processing nodes of a parallel (shared or distributed memory) computer or distributed system is more promising t,oclay than at any time in history of the field. Accelerating simulation experiments for la.rge syst,em models is naturally the outstanding PDS resea.rch goal. But above this, many scientists, also from within the core of classical PDS field, have seen imp& of the PDS theory and methodology also in nonsta.nda.rd a.pplication areas such as parallel program execution. The intent, of this minitrack was to provide a forum for t,he exploration of new high performance simula.tion concepts and techniques, as well as their successful a.pplication on today’s and tomorrow’s parallel execution pla.tforms. Twenty-six contributions have been submitted by a.utjhors from France, Germany, Japa.n, Poland, Sweden, UK and the USA, after more than seventy authors a~nnouacecl pa.pers via abstracts. Eighty-four reviewers wit#h their expertise helped to select ten full papers a.nd two short papers for publication in the proceedings, a,ncl for oral presentation at the conference. Ea.& paper was reviewed by at least three, at most, six, and on a,verage by 3.96 reviewers. Due to t)he a.va.ilability of pa.pers in electronic format, the reviewing process could be managed almost exclusively by email. A concept8ual framework and an environment for high performance simulation is presented in the first paper. Zeigler, Moon, D. Kim and J.G. Kim propose the DEVS formalism for system modeling, and a DEVS implementation based on object oriented technology enabling a full exploita,tion of a para.llel simulation model execution. DEVS modeling is illustra.ted in the context of wildland fire simulation, performance comparisons a.re conducted for a, watershed model executing sequentially on a. Spare-1000 a.ncl in para.llel on a CM-5. The paper by Konas presents potentials of simulation a.t the confluence of object oriented system design and parallel processing: modularity, extensibility and reusability provide a na.tural a.nd well structured approach to the construction of complex simulation models, while at the same time promoting a higher execution efficiency on a. parallel platform. Young and Wilsey propose a new distributed fossil collection technique for the Time Wa.rp distributed discrete event simulation protocol. Basically, every fossil collector associated with a logical process (LP) by observing event arrival t imes establishes a. statistical model for rollback distances to determine in conjunction with a user defined rise fa.ctor tha.t controls the aggressiveness of fossil collectors a. probabilistic GVT bound. The approach appears beneficial for the reduction of the amount of used memory over GVT based fossil collection, but requires a.dditional checkpointing for possible “ca.tastrophic” rollbacks, i.e. restoration of states that have been fossil collected due to an overestimation of the actual GVT. RGnngren, Barriga a,nd Ayani have developed a benchmark suite for the performance evalua.tion of parallel simulation kernels on different a.rchitect,ures and the scalability analysis of certain simulation problems. It appears particularly hard t,o isolate performance influences stemming from the kernel a.s such or from the event structure underlying the simulation model executed by the kernel. Trying to a,bstra,ct a.s much a,s possible from the la.tter performance impact, the authors construct a. synthetic benchmark scala,ble

Cube-4—a scalable architecture for real-time volume rendering

Abstract: We present Cube-4, a special-purpose volume rendering architecture that is capable of rendering high-resolution (e.g., 1024/sup 3/) datasets at 30 frames per second. The underlying algorithm, called slice-parallel ray-casting, uses tri-linear interpolation of samples between data slices for parallel and perspective projections. The architecture uses a distributed interleaved memory, several parallel processing pipelines, and an innovative parallel data flow scheme that requires no global communication, except at the pixel level. This leads to local, fixed bandwidth interconnections and has the benefits of high memory bandwidth, real-time data input, modularity, and scalability. We have simulated the architecture and have implemented a working prototype of the complete hardware on a configurable custom hardware machine. Our results indicate true real-time performance for high-resolution datasets and linear scalability of performance with the number of processing pipelines.

Scalable compression and transmission of internet multicast video

Abstract: In just a few years the "Internet Multicast Backbone", or MBone, has risen from a small, research curiosity to a large scale and widely used communications infrastructure. A driving force behind this growth was our development of multipoint audio, video, and shared whiteboard conferencing applications that are now used daily by the large and growing MBone community. Because these real-time media are transmitted at a uniform rate to all the receivers in the network, the source must either run below the bottleneck rate or overload portions of the multicast distribution tree. In this dissertation, we propose a solution to this problem by moving the burden of rate-adaptation from the source to the receivers with a scheme we call Receiver-driven Layered Multicast, or RLM. In RLM, a source distributes a hierarchical signal by striping the constituent layers across multiple multicast groups. Receivers then adjust their reception rate by simply joining and leaving multicast groups. But RLM solves only half of the problem. To distribute a multi-rate flow to heterogeneous receivers using RLM, the underlying signal must be encoded in a hierarchical or layered format. To this end, we developed and present herein a layered video compression algorithm which, when combined with RLM, provides a comprehensive solution for scalable multicast video transmission in heterogeneous networks. In addition to a layered representation, our coder has low-complexity (admitting an efficient software implementation) and high error resilience (admitting robust operation in loosely controlled environments like the Internet). Our RLM/PVH framework leverages two design methodologies from two related yet often segregated fields: joint source/channel coding (JSCC) from traditional communications theory and application level framing (ALF) from computer network design. In accordance with JSCC, we combine the design of the source-coding algorithm (i.e., PVH) with the channel-coding algorithm (i.e., RLM), while in accordance with ALF, we reflect application semantics (i.e., PVH) in the design of the network protocol (i.e., RLM). In this thesis, we posit that JSSC and ALF are two manifestations of the same underlying design principle. We explore the ALF/JSSC design space with a discussion of our "Intra-H.261" video coder, which we developed specifically for MBone video transmission, and compare its performance to that of traditional designs based on independent source- and channel-coding. Finally, we bring all of the pieces of our design together into a comprehensive system architecture realized in a flexible software toolkit that underlies our widely used video application--the UCB/LBL video conferencing tool vic. Our system architecture not only integrates RLM and PVH into an autonomous video application but also provides the functionality requisite to a complete multimedia communication system, including user-interface elements and companion applications like audio and shared whiteboard. In this framework, we craft "media agents" from a common multimedia toolkit and control and configure them over a software interprocess communication bus that we call the Coordination Bus. By composing an arbitrary arrangement of media agents over the Coordination Bus and complementing the arrangement with an appropriate user-interface, we can induce an arbitrary multimedia collaboration style. Unlike previous work on layered video compression and transmission, we have implemented RLM, PVH, and our coordination framework in a "real" application and are deploying a fully operational system on a very large scale over the MBone. (Abstract shortened by UMI.)

Network topologies for scalable multi-user virtual environments

Abstract: Investigates the trade-off of different network topologies and messaging protocols for multi-user virtual environment systems. We present message distribution techniques appropriate for constructing scalable multi-user systems for a variety of network characteristics. Hierarchical system designs utilizing servers that manage message distribution for entities in separate regions of a virtual environment are described that scale to arbitrary numbers of simultaneous users. Experimental results show that the rate of messages processed by server workstations in this system design are less than using previously described approaches.

Software infrastructure for the I-WAY high-performance distributed computing experiment

Abstract: High speed wide area networks are expected to enable innovative applications that integrate geographically distributed, high performance computing, database, graphics, and networking resources. However, there is as yet little understanding of the higher level services required to support these applications, or of the techniques required to implement these services in a scalable, secure manner. We report on a large scale prototyping effort that has yielded some insights into these issues. Building on the hardware base provided by the I-WAY, a national scale asynchronous transfer mode (ATM) network, we developed an integrated management and application programming system, called I-Soft. This system was deployed at most of the 17 I-WAY sites and used by many of the 60 applications demonstrated on the I-WAY network. We describe the I-Soft design and report on lessons learned from application experiments.

Making World Wide Web Caching Servers Cooperate.

Abstract: Due to its exponential growth, the World Wide Web is increasingly experiencing several problems, such as hot spots, increased network bandwidth usage, and excessive document retrieval latency The standard solution to these problems is to use a caching proxy However, a single caching proxy is a bottleneck: there is a limit to the number of clients that can use the same cache, and thereby the effectiveness of the cache is limited Also, a caching proxy is a single point of failure We address these problems by creating a protocol that allows multiple caching proxies to cooperate and share their caches, thus increasing robustness and scalability This scalability, in turn, gives each client an effectively larger cache with a higher hit rate This paper describes a prototype implementation of this protocol that uses IP multicast to communicate between the servers

A Case for Network-Attached Secure Disks,

Abstract: : By providing direct data transfer between storage and client, network-attached storage devices have the potential to improve scalability (by removing the server as a bottleneck) and performance (through network striping and shorter data paths). Realizing the technology's full potential requires careful consideration across a wide range of file system, networking and security issues. To address these issues, this paper presents two new network-attached storage architectures. (1) Networked SCSI disks (NetSCSI) are network-attached storage devices with minimal changes from the familiar SCSI interface (2) Net-work-attached secure disks (NASD) are drives that support independent client access to drive provided object services. For both architectures, we present a sketch of repartitionings of distributed file system functionality, including a security framework whose strongest levels use tamper-resistant processing in the disks to provide action authorization and data privacy even when the drive is in an physically insecure location.

Power Load Management as a Computational Market

Abstract: Power load management enables energy utilities to reduce peak loads and thereby save money. Due to the large number of different loads, power load management is a complicated optimization problem. We present a new decentralized approach to this problem by modeling direct load management as a computational market. Our simulation results demonstrate that our approach is very efficient with a superlinear rate of convergence to equilibrium and an excellent scalability, requiring few iterations even when the number of agents is in the order of one thousand. A framework for analysis of this and similar problems is given which shows how nonlinear optimization and numerical mathematics can be exploited to characterize, compare and tailor problem-solving strategies in market-oriented programming.

The HyperDisco approach to open hypermedia systems

Abstract: Computing support for large engineering enterprises provides an example of the need for hypermedia-based collaborative computing systems composed of a large number of distributed heterogeneous tools. These computing environments place complex requirements on the underlying hypermedia platform. To support integration of independently written tools for these environments, hypermedia platforms must address several important issues such as scalability, openness, distribution, heterogeneity, interoperability, extensibility and computation. This paper describes the HyperDisco approach to open hypermedia systems. HyperDisco provides an extensible object-oriented hypermedia platform supporting intertool linking, computation, concurrency control, noti cation control, version control, access control, query and search, and various other features. The present work has two main objectives: 1) to provide a platform to integrate existing and future distributed heterogeneous tools and data formats and 2) to provide a platform to extend integrated tools to handle multiple collaborating users and multiple versions of shared artifacts. The paper presents important dimensions of hypermedia platforms that helped to formulate the goals for HyperDisco, the HyperDisco prototype, and two integration examples to illustrate the distinctive features of the HyperDisco approach.

3D geographic network displays

Abstract: Many types of information may be represented as graphs or networks with the nodes corresponding to entities and the links to relationships between entities. Often there is geographical information associated with the network. The traditional way to visualize geographical networks employs node and link displays on a two-dimensional map. These displays are easily overwhelmed, and for large networks become visually cluttered and confusing. To overcome these problems we have invented five novel network views that generalize the traditional displays. Two of the views show the complete network, while the other three concentrate on a portion of a larger network defined by connectivity to a given node. Our new visual metaphors retain many of the well-known advantages of the traditional network maps, while exploiting three-dimensional graphics to address some of the fundamental problems limiting the scalability of two-dimensional displays.

Bayou: replicated database services for world-wide applications

Abstract: The Bayou architecture provides scalability, availability, extensibility, and adaptability features that address database storage needs of world-wide applications. In addition to discussing these features, this paper presents Bayou's mechanisms for permitting the replicas of a database to vary dynamically without global coordination. Key is the use of weak consistency replication among autonomous machines and strict adherence to the tenet that no operation should involve more than two machines.

An evaluation of the Amoeba group communication system

Abstract: The Amoeba group communication system has two unique aspects: (1) it uses a sequencer-based protocol with negative acknowledgements for achieving a total order on all group messages; and (2) users choose the degree of fault tolerance they desire. This paper reports on our design decisions in retrospect, the performance of the Amoeba group system, and our experiences using the system. We conclude that sequencer-based group protocols achieve high performance (comparable to Amoeba's fast remote procedure call implementation), that the scalability of our sequencer-based protocols is limited by message processing time, and that the flexibility and modularity of user-level implementations of protocols is likely to outweigh the potential performance loss.

An extensible meta-learning approach for scalable and accurate inductive learning

Abstract: Much of the research in inductive learning concentrates on problems with relatively small amounts of data. With the coming age of ubiquitous network computing, it is likely that orders of magnitude more data in databases will be available for various learning problems of real world importance. Some learning algorithms assume that the entire data set fits into main memory, which is not feasible for massive amounts of data, especially for applications in data mining. One approach to handling a large data set is to partition the data set into subsets, run the learning algorithm on each of the subsets, and combine the results. Moreover, data can be inherently distributed across multiple sites on the network and merging all the data in one location can be expensive or prohibitive. In this thesis we propose, investigate, and evaluate a meta-learning approach to integrating the results of multiple learning processes. Our approach utilizes machine learning to guide the integration. We identified two main meta-learning strategies: combiner and arbiter. Both strategies are independent to the learning algorithms used in generating the classifiers. The combiner strategy attempts to reveal relationships among the learned classifiers' prediction patterns. The arbiter strategy tries to determine the correct prediction when the classifiers have different opinions. Various schemes under these two strategies have been developed. Empirical results show that our schemes can obtain accurate classifiers from inaccurate classifiers trained from data subsets. We also implemented and analyzed the schemes in a parallel and distributed environment to demonstrate their scalability.

Connectivity and sparse wavelength conversion in wavelength-routing networks

Abstract: Wavelength-routing networks offer the advantages of wavelength re-use and scalability over broadcast-and-select networks and are therefore suitable for wide area networks (WANs). We study the effects of topological connectivity and wavelength conversion in circuit-switched all-optical wavelength-routing networks. An approximate blocking analysis of such network is performed. We first propose an improved framework for the analysis of networks with arbitrary topology. We introduce a simple model for networks with a variable number of converters and analyze the effect of wavelength converter density on blocking probability. We then apply this framework to two sparse network topologies, the ring and the mesh-torus, and obtain the blocking performance. The results show that, in most cases, only a fraction of the network nodes need to be equipped with wavelength conversion capability for good performance. Finally, the tradeoff between physical connectivity, wavelength conversion, and the number of available wavelengths is studied through networks with random topologies.

Design and performance tradeoffs in clustered video servers

Abstract: We investigate the suitability of clustered architectures for designing scalable multimedia servers. Specifically, we evaluate the effects of: (i) architectural design of the cluster; (ii) the size of the unit of data interleaving; and (iii) read ahead buffering and scheduling on the real time performance guarantees provided by the server. To analyze the effects of these parameters, we develop an analytical model of clustered multimedia servers, and then validate it through extensive simulations, The results of our analysis have formed the basis of our prototype implementation based on an RS/6000 Scalable Parallel (SP) machine. We briefly describe the prototype and discuss some implementation details.

Code generation for transport triggered architectures

Abstract: Transport triggered architectures (TTAs) form a class of architectures which are programmed by specifying data transports between function units. As side effect of these data transports these function units perform operations. Making these data transports visible at the architectural level contributes to the flexibility and scalability of processors. Furthermore it enables several extra code scheduling optimizations. These properties make TTAs very suitable for being applied for embedded processors.

The Hyper-G Network Information System

Abstract: As the Internet continues to experience exponential rates of growth, attention is shifting away from mainstream network services such as electronic mail and file transfer to more interactive information services. Current network information systems, whilst extremely successful, run into problems of fragmentation, consistency, scalability, and loss of orientation.

Efficient and flexible location management techniques for wireless communication systems

Abstract: We consider the problem of managing the information required to locate users in a wireless communication system, with a focus on designing and evaluating location management techniques that are efficient, scalable, and flexible. The three key contributions of this paper are: (1) a family of location management techniques, HiPER (for Hierarchical ProfilE Replication), that efficiently provide life‐long (non‐geographic) numbering with fast location lookup; (2) Pleiades, a scalable event‐driven wireless system simulator with realistic calling and mobility patterns derived from several months of real traffic traces; and (3) multi‐day simulations comparing our proposed location management techniques with current and previously proposed techniques on a realistic geographical and network topology.

Global communication analysis and optimization

Abstract: Reducing communication cost is crucial to achieving good performance on scalable parallel machines. This paper presents a new compiler algorithm for global analysis and optimization of communication in data-parallel programs. Our algorithm is distinct from existing approaches in that rather than handling loop-nests and array references one by one, it considers all communication in a procedure and their interactions under different placements before making a final decision on the placement of any communication. It exploits the flexibility resulting from this advanced analysis to eliminate redundancy, reduce the number of messages, and reduce contention for cache and communication buffers, all in a unified framework. In contrast, single loop-nest analysis often retains redundant communication, and more aggressive dataflow analysis on array sections can generate too many messages or cache and buffer contention. The algorithm has been implemented in the IBM pHPF compiler for High Performance Fortran. During compilation, the number of messages per processor goes down by as much as a factor of nine for some HPF programs. We present performance results for the IBM SP2 and a network of Sparc workstations (NOW) connected by a Myrinet switch. In many cases, the communication cost is reduced by a factor of two.