We propose the Geometric Bounds (GBs), a new family of fast and accurate noniterative bounds on closed queueing network performance metrics that can be used in the online optimization of distributed applications. Compared to state-of-the-art techniques such as the Balanced Job Bounds (BJBs), GB achieves higher accuracy at similar computational costs, limiting the worst- case bounding error typically within 5-13 percent when, for the BJB, it is usually in the range of 15-35 percent. Optimization problems that are solved with GBs return solutions that are much closer to the global optimum than with existing bounds. We also show that the GB technique generalizes as an accurate approximation to closed fork-join networks commonly used in disk, parallel, and database models, thus extending the applicability of the method beyond the optimization of basic product-form networks.

Routing algorithms such as Distance Vector and Link States have the routing table size as Omega (n), where n is the number of destination identifiers, thus providing only limited scalability for large networks when n is high. As the distributed hash table (DHT) techniques are extraordinarily scalable with n, our work aims at adapting a DHT approach to the design of a network- layer routing algorithm so that the average routing table size can be significantly reduced to O (log n) without losing much routing efficiency. Nonetheless, this scheme requires a major breakthrough to address some fundamental challenges. Specifically, unlike a DHT, a network-layer routing algorithm must (1) exchange its control messages without an underlying network, (2) handle link insertion/deletion and link-cost updates, and (3) provide routing efficiency. Thus, we are motivated to propose a new network-layer routing algorithm, Tunnel Vector (TV), using DHT-like multilevel routing without an underlying network. TV exchanges its control messages only via physical links and is self-configurable in response to linkage updates. In TV, the routing path of a packet is near optimal while the routing table size is O(log n) per node, with high probability. Thus, TV is suitable for routing in a very large network.

A new emerging class of problems related to the online configuration and optimization of computer systems and networks requires the solution in a very short amount of time of a large number of analytical performance models, often based on queueing networks. In this paper we propose the Geometric Bounds (GB), a new family of fast noniterative bounds on performance metrics of closed productform queueing networks. In spite of their simplicity, the proposed bounds are more accurate than the popular Balanced Job Bounds (BJB), even in the difficult case of networks with multiple bottlenecks or large delays.

This paper considers the problem of mining closed frequent itemsets over a data stream sliding window using limited memory space. We design a synopsis data structure to monitor transactions in the sliding window so that we can output the current closed frequent itemsets at any time. Due to time and memory constraints, the synopsis data structure cannot monitor all possible itemsets. However, monitoring only frequent itemsets will make it impossible to detect new itemsets when they become frequent. In this paper, we introduce a compact data structure, the closed enumeration tree (CET), to maintain a dynamically selected set of itemsets over a sliding window. The selected itemsets contain a boundary between closed frequent itemsets and the rest of the itemsets. Concept drifts in a data stream are reflected by boundary movements in the CET. In other words, a status change of any itemset (e.g., from non-frequent to frequent) must occur through the boundary. Because the boundary is relatively stable, the cost of mining closed frequent itemsets over a sliding window is dramatically reduced to that of mining transactions that can possibly cause boundary movements in the CET. Our experiments show that our algorithm performs much better than representative algorithms for the sate-of-the-art approaches.

We propose a new overlay network, called Generic Identifier Network (GIN), for collaborative nodes to share objects with transactions across affiliated organizations by merging the organizational local namespaces upon mutual agreement. Using local namespaces instead of a global namespace can avoid excessive dissemination of organizational information, reduce maintenance costs, and improve robustness against external security attacks. GIN can forward a query with an O(1) latency stretch with high probability and achieve high performance. In the absence of a complete distance map, its heuristic algorithms for self configuration are scalable and efficient. Routing tables are maintained using soft-state mechanisms for fault tolerance and adapting to performance updates of network distances. Thus, GIN has significant new advantages for building an efficient and scalable distributed hash table for modern collaborative applications across organizations.

We consider the problem of resource allocation in mining multiple data streams. Due to the large volume and the high speed of streaming data, mining algorithms must cope with the effects of system overload. How to realize maximum mining benefits under resource constraints becomes a challenging task. In this paper, we propose a load shedding scheme for classifying multiple data streams. We focus on the following problems: i) how to classify data that are dropped by the load shedding scheme? and ii) how to decide when to drop data from a stream? We introduce a quality of decision (QoD) metric to measure the level of uncertainty in classification when exact feature values of the data are not available because of load shedding. A Markov model is used to predict the distribution of feature values and we make classification decisions using the predicted values and the QoD metric. Thus, resources are allocated among multiple data streams to maximize the quality of classification decisions. Furthermore, our load shedding scheme is able to learn and adapt to changing data characteristics in the data streams. Experiments on both synthetic data and real-life data show that our load shedding scheme is effective in improving the overall accuracy of classification under resource constraints.

Tree structures are used extensively in domains such as computational biology, pattern recognition, XML databases, computer networks, and so on. One important problem in mining databases of trees is to find frequently occurring subtrees. Because of the combinatorial explosion, the number of frequent subtrees usually grows exponentially with the size of frequent subtrees and, therefore, mining all frequent subtrees becomes infeasible for large tree sizes. We present CMTreeMiner, a computationally efficient algorithm that discovers only closed and maximal frequent subtrees in a database of labeled rooted trees, where the rooted trees can be either ordered or unordered. The algorithm mines both closed and maximal frequent subtrees by traversing an enumeration tree that systematically enumerates all frequent subtrees. Several techniques are proposed to prune the branches of the enumeration tree that do not correspond to closed or maximal frequent subtrees. Heuristic techniques are used to arrange the order of computation so that relatively expensive computation is avoided as much as possible. We study the performance of our algorithm through extensive experiments, using both synthetic data and data sets from real applications. The experimental results show that our algorithm is very efficient in reducing the search space and quickly discovers all closed and maximal frequent subtrees.

This paper considers the problem of mining closed frequent itemsets over a sliding window using limited memory space. We design a synopsis data structure to monitor transactions in the sliding window so that we can output the current closed frequent itemsets at any time. Due to time and memory constraints, the synopsis data structure cannot monitor all possible itemsets. However, monitoring only frequent itemsets make it impossible to detect new itemsets when they become frequent. In this paper, we introduce a compact data structure, the closed enumeration tree (CET), to maintain a dynamically selected set of item-sets over a sliding-window. The selected itemsets consist of a boundary between closed frequent itemsets and the rest of the itemsets. Concept drifts in a data stream are reflected by boundary movements in the CET. In other words, a status change of any itemset (e.g., from non-frequent to frequent) must occur through the boundary. Because the boundary is relatively stable, the cost of mining closed frequent item-sets over a sliding window is dramatically reduced to that of mining transactions that can possibly cause boundary movements in the CET. Our experiments show that our algorithm performs much better than previous approaches.

The World Wide Web is the largest publicly available image repository and a natural source of attention. An immediate consequence is that searching for images on the Web has become a current and important task. To search for images of interest, the most direct approach is keyword-based searching. However, since images on the Web are poorly labeled, direct application of standard keyword-based image searching techniques frequently yields poor results. We propose a comprehensive solution to this problem. In our approach, multiple sources of evidence related to the images are considered. To allow combining these distinct sources of evidence, we introduce an image retrieval model based on Bayesian belief networks. To evaluate our approach, we perform experiments on a reference collection composed of 54000 Web images. Our results indicate that retrieval using an image surrounding text passages is as effective as standard retrieval based on HTML tags. This is an interesting result because current image search engines in the Web usually do not take text passages into consideration. Most important, according to our results, the combination of information derived from text passages with information derived from HTML tags leads to improved retrieval, with relative gains in average precision figures of roughly 50 percent, when compared to the results obtained by the use of each source of evidence in isolation.

Tree structures are used extensively in domains such as computational biology, pattern recognition, computer networks, and so on. We present an indexing technique for free trees and apply this indexing technique to the problem of mining frequent subtrees. We first define a novel representation, the canonical form, for rooted trees and extend the definition to free trees. We also introduce another concept, the canonical string, as a simpler representation for free trees in their canonical forms. We then apply our tree indexing technique to the frequent subtree mining problem and present FreeTreeMiner, a computationally efficient algorithm that discovers all frequently occurring subtrees in a database of free trees. We study the performance and the scalability of our algorithms through extensive experiments based on both synthetic data and datasets from two real applications: a dataset of chemical compounds and a dataset of Internet multicast trees.

In this paper, we investigate in the problem of efficient filtering of a large dataset according to some user specified criterion. A novel paradigm is introduced to allow users to easily integrate their prior knowledge in the mining process and to customize the mining result. The Bayesian Network is chosen as the model to represent user's knowledge due to its rich expressive power, compact representation, and intuitive semantics. However, one difficulty that prevents the Bayesian network model from being applied to large scale databases in a brute-force way is the generally high computational complexity of the inference algorithms. To tackle this challenge, we propose three techniques, namely network pruning of conditional probability, computation reuse by tuple reordering, and early termination, to speed up the process dramatically. It has been demonstrated that our proposed algorithm can save as much as 95% of the execution time empirically.

Data mining has been an area of increasing interest. The association rule discovery problem in particular has been widely studied. However, there are still some unresolved problems. For example, research on mining patterns in the evolution of numerical attributes is still lacking. This is both a challenging problem and one with significant practical applications in business, science, and medicine. In this paper we present a temporal association rule model for evolving numerical attributes. Metrics for qualifying a temporal association rule include the familiar measures of support and strength used in traditional association rule mining and a new metric called density. The density metric not only gives us a way to extract the rules that best represent the data, but also provides an effective mechanism to prune the search space. An efficient algorithm is devised for mining temporal association rules, which utilizes all three thresholds (especially the strength) to prune the search space drastically. Moreover, the resulting rules are represented in a concise manner via rule sets to reduce the output size. Experimental results on real and synthetic data sets demonstrate the efficiency of our algorithm.

The bulk of proposed pervasive computing devices such as PDAs and cellular telephones operate as thin clients within a larger infrastructure. To access services within their local environment, these devices participate in a service discovery protocol which involves a master directory that registers all services available in the local environment. These directories typically are isolated from each other. Devices that move across service discovery domains have no access to information outside their current local domain. In this paper we propose an application-level protocol called VIA that enables data sharing among discovery domains. Each directory maintains a table of active links to other directories that share related information. A set of linked directories forms a data cluster that can be queried by devices for information. The data cluster is distributed, self-organizing, responsive to data mobility, and robust to failures. Using application-defined data schemas, clusters organize themselves into a hierarchy for efficient querying and network resource usage. Through analysis and simulation we describe the behavior of VIA under different workloads and show that the protocol overhead for both maintaining a cluster and handling failures grows slowly with the number of gateways.

Despite enormous progress in networking and computing technologies, their application has remained restricted to conventional person-to-person and person-to-computer communication. However, continual reduction in cost and form factor is now making it possible to imbed networking - even wireless networking - and computing capabilities not just in our PCs and laptops but also other objects. Further, a marriage of these ever tinier and cheaper processors and wireless network interfaces with emerging micro-sensors based on MEMS technology is allowing cheap sensing, processing, and communication capabilities to be unobtrusively embedded in familiar physical objects. The result is an emerging paradigm shift where the primary role of information technology would be to enhance or assist in "person to physical world communication via familiar physical objects with embedded (a) micro-sensors to react to external stimuli, and (b) wireless networking and computing engines for tetherless communication with compute servers and other networked embedded objects. In this paper we present the application of sensor-based wireless networks to a "Smart Kindergarten that we are developing to target developmental problem-solving environments for early childhood education. This is a natural application as young children learn by exploring and interacting with objects such as toys in their environment. Our envisioned system would enhance the education process by providing a childhood learning environment that is individualized to each child, adapts to the context, coordinates activities of multiple children, and allows continual unobtrusive evaluation of the learning process by the teacher. This would be done by wirelessly-networked, sensor-enhanced toys and other classroom objects with back-end middleware services and database techniques. We explore wireless networking, middleware, and data management technologies for realizing this application, and describe challenges arising from ad hoc distributed structure, unreliable sensing, large scale/density, and novel sensor data types that are characteristic of such deeply instrumented environments with inter-networked physical objects.

Recent technological advances in digital signal processing, data compression techniques, and high-speed communication networks have made Video-on-Demand (VOD) servers feasible. A challenging task in such systems is servicing multiple clients simultaneously while satisfying real-time requirements of continuous delivery of objects at specified rates. To accomplish these tasks and realize economies of scale associated with servicing a large user population, a VOD server requires a large disk subsystem. Although a single disk is fairly reliable, a large disk farm can have an unacceptably high probability of disk failure. Furthermore, due to real-time constraints, the reliability requirements of VOD systems are even more stringent than those of traditional information systems. Traditional RAID solutions are inadequate due to poor resource usage. Thus, in this paper, we present alternative schemes which provide a high degree of reliability at low disk storage, bandwidth, and memory costs for on-demand multimedia servers. Moreover, we discuss some of the main issues and trade-offs associated with providing fault tolerance in multidisk VOD systems. We would like to impress upon the reader that one of the main points of this paper is the exposition of trade-offs and issues associated with designing fault-tolerant VOD servers. It is not the case that one fault tolerance scheme is absolutely better than another, but rather that one must understand the trade-offs as well as one's system constraints and then choose a fault tolerance scheme accordingly.

In light of advances in processor and networking technology, especially the emergence of network attached disks, the traditional client-server architecture of file systems has become suboptimal for many computation/data intensive applications. We introduce a revised architecture for file management employing network attached storage: the dynamic file server environment (Dynamo). Dynamo introduces two main architectural innovations: to provide high scalability, the file management functions are mainly performed cooperatively by the clients in the system. Furthermore, data is transferred directly to the client's cache from network-attached disks, thus avoiding copies from a disk to the server buffer and then over the network to the client. Secondly, Dynamo uses a cooperative cache management which employs a decentralized lottery-based page replacement strategy. We show via performance benchmarks run on the Dynamo system and simulation results how this architecture increases the system's adaptability, scalability and cost performance.

The concept of time and the ordering of events are correlated key issues in distributed computing as well as in computer-supported collaborative work (CSCW) systems. The paper revisits some of the assumptions that have generally been made regarding time and event ordering in distributed systems and argues that they are no longer appropriate if the goal is to faithfully preserve user intentions in CSCW systems. In particular, the following contributions are made in the context of collaborative editing systems. First, we discuss how the user intentions might be impacted when the finite duration of drawing operations are considered. Secondly, we propose that the total ordering of events should give the users the right of participation instead of being solely determined mechanically by the system. Thirdly, a new concept of an active whiteboard is proposed which supports various integrity constraints on objects and object groups to maintain user intentions in a more sophisticated way. Additionally, for the sake of completeness, the problem of maintaining consistency in the face of unreliable and high-latency communication channels is also covered.

We compare performance of a multimedia storage server based on a random data allocation layout and block replication with traditional data striping techniques. Data striping techniques in multimedia servers are often designed for restricted workloads, e.g. sequential access patterns with CBR (constant bit rate) requirements. On the other hand, a system based on random data allocation can support virtually any type of multimedia application, including VBR (variable bit rate) video or audio, and interactive applications with unpredictable access patterns, such as 3D interactive virtual worlds, interactive scientific visualizations, etc. Surprisingly, our results show that system performance with random data allocation is competitive and sometimes even outperforms traditional data striping techniques, for the workloads for which data striping is designed to work best; i.e. streams with sequential access patterns and CBR requirements. Due to its superiority in supporting general workloads and competitive system performance, we believe that random data allocation will be the scheme of choice for next generation multimedia servers.

No abstract available.

With the exponential growth of hosts and traffic workloads on the Internet, collaborative web caching has been recognized as an efficient solution to alleviate web page server bottlenecks and reduce traffic. However, cache discovery, i.e., locating where a page is cached, is a challenging problem, especially in the fast growing World Wide Web environment, where the number of participating proxies can be very large. In this paper, we propose a new scheme which employs proxy affinities to maintain a dynamic distributed collaborative caching infrastructure. Web pages are partitioned into clusters according to proxy reference patterns. All proxies which frequently access some page(s) in the same web page cluster form an “information group”. When web pages belonging to a web page cluster are deleted from or added into a proxy's cache, only proxies in the associated information group are notified. This scheme can be shown to greatly reduce the number of messages and other overhead on individual proxies while maintaining a high cache hit rate. Finally, we employ trace driven simulation to evaluate our web caching scheme using three web access trace logs to verify that our caching structure can provide significant benefits on real workloads.

In light of advances in processor and networking technology, especially the emergence of network attached disks, the traditional client-server architecture of persistent storage systems has become suboptimal for many computation/data intensive applications. In this paper, we introduce a revised architecture for persistent object management employing network attached storage: the dynamic object server environment (Dynamo). Dynamo introduces two main architectural innovations: (1) to provide high scalability, the object management functions are mainly performed cooperatively by the clients in the system. Furthermore, data is transferred directly to the client's cache from network-attached disks, thus avoiding copies from a disk to the server buffer and then over the network to the client. Other systems have an a priori assignment of ‘object management’ functions to specific nodes; (2) Dynamo uses a cooperative cache management which employs a decentralized lottery-based page replacement strategy with a novel technique which attempts to put a page replaced from one node into the cache of another node which is likely to access the page in the near future. We show via performance benchmarks run on the Dynamo system and simulation results how this architecture increases the system's adaptability, scalability and cost performance. Copyright © 2000 John Wiley & Sons, Ltd.

Spatial data mining presents new challenges due to the large size of spatial data, the complexity of spatial data types, and the special nature of spatial access methods. Most research in this area has focused on efficient query processing of static data. This paper introduces an active spatial data mining approach that extends the current spatial data mining algorithms to efficiently support user-defined triggers on dynamically evolving spatial data. To exploit the locality of the effect of an update and the nature of spatial data, we employ a hierarchical structure with associated statistical information at the various levels of the hierarchy and decompose the user-defined trigger into a set of subtriggers associated with cells in the hierarchy. Updates are suspended in the hierarchy until their cumulative effect might cause the trigger to fire. It is shown that this approach achieves three orders of magnitude improvement over the naive approach that reevaluate the condition over the database for each update, while both approaches produce the same result without any delay. Moreover, this scheme can support incremental query processing as well.

The importance of supporting flexible roles and dynamic policies has long been recognized in the CSCW (Computer-Supported Cooperative Work) literature but unfortunately never with a systematic solution. This paper proposes a taxonomy of runtime dynamics in collaborative systems in general and discusses our solution in the COCA framework. Firstly, individual participants can take roles, drop roles, and switch between roles as a collaboration runs, Secondly, a role can be transferred from one participant to another at runtime. Thirdly, session-wide, authorized participants can change the definition of roles and coordination policies on the fly while keeping the mapping between roles and participants. When such changes happen on the fly, the runtime environment of the sites which are potentially affected, including the state information within the computation modules and the communication channels, must be handled properly according to the semantics of the collaboration itself.

Extensibility and high query performance are important requirements of advanced large scale information systems since complex data analysis often requires the use of application-specific operations that have to be introduced by the user issuing the query. Towards the goal of supporting automatic parallelization of queries containing complex user-defined evaluators in an extensible DBMS, we devised a relevance window model to capture the inherent data lineage characteristics of evaluators on multidimensional data sets. Informally, the relevance window of an evaluator defines the scope of influence input data records have on the value of records in the output data space. An evaluator's relevance window constrains the data partitioning opportunities available for an evaluator.

Spatial data mining presents new challenges due to the large size of spatial data, the complexity of spatial data types, and the special nature of spatial access methods. Most research in this area has focused on efficient query processing of static data. This paper introduces an active spatial data mining approach which extends the current spatial data mining algorithms to efficiently support user-defined triggers on dynamically evolving spatial data. To exploit the locality of the effect of an update and the nature of spatial data, we employ a hierarchical structure with associated statistical information at the various levels of the hierarchy and decompose the user-defined trigger into a set of sub-triggers associated with cells in the hierarchy. Updates are suspended in the hierarchy until their cumulative effect might cause the trigger to fire. It is shown that this approach achieves three orders of magnitude improvement over the naive approach that re-evaluates the condition over the database for each update, while both approaches produce the same result without any delay. Moreover this scheme can support incremental query processing as well.

Full support of parallelism in object-relational database systems (ORDBMSs) is desired. The parallelization techniques developed for relational database systems are not adequate for ORDBMS because of the introduction of complex abstract data types and operations on ordered domains. In this paper, we consider a data stream paradigm and develop a query parallelization framework that exploits characteristics of user-defined functions in a ORDBMS during query optimization. By introducing the concept of windows and abstract data type orderings, we develop a novel approach for parallelizing user-defined functions in a distributed ORDBMS environment. The implementation issues in providing query services in ordered domains are also discussed.

Very long running queries in database systems are not uncommon in non traditional application domains such as image processing or data warehousing analysis. Query optimization, therefore, is important. However, estimates of the query characteristics before query execution are usually inaccurate. Further, system configuration and resource availability may change during long evaluation period. As a result, queries are often evaluated with sub-optimal plan configurations. To remedy this situation, we have designed a novel approach to re-optimize suboptimal query plan configurations on-the-fly with Conquest, an extensible and distributed query processing system. A dynamic optimizer considers reconfiguration cost as well as execution cost in determining the best query plan configuration. Experimental results are presented.

Coordination poticies vary from collaboration to collaboration and are even subject to evolution in different phases of the same co~aboration.It is titi for co~aborative systems to be flexible enough to accommodate changes to the coordination poficies during development and the Kietime of the coHaboration.Motivated by previous work of separating coordination and computation, we propose COCA as a generic &amew'ork for developing co~aborative systems and modeling the coordination poficies.Participants expficidy adopt roles, and coordination poficies are sptid in terms of roIes in a logic-based specification language.Po5cies me interpret at runtime at each collaboration site by a COCA virturd machine.It is easy to change the coordination policies both during development and at runtime.KeSvor&

Massive database sizes and growing demands for decision support and data mining result in long-running queries in extensible object-relational DBMSs, particularly in decision support and data warehousing analysis applications. Parallelization of query evaluation is often required for acceptable performance, yet queries are frequently processed suboptimally due to (1) only coarse or inaccurate estimates of the query characteristics and database statistics being available prior to query evaluation; (2) changes in system configuration and resource availability during query evaluation. In a distributed environment, dynamically reconfiguring query execution plans (QEPs), which adapts QEPs to the environment as well as to the query characteristics, is a promising means to significantly improve query evaluation performance. Based on an operator classification, we propose an algorithm to coordinate the steps in a reconfiguration and introduce alternatives for execution context checkpointing and restoring. A syntactic extension of SQL to expose the relevant characteristics of user-defined functions in support of dynamic reconfiguration is proposed. An example from the experimental system is presented.

HO is a mtdtimedia object server which manages a set of parfld disks and supports red-time data d&very tith statisticrd dday guarantees.MO uses random data aRocation on diik< combiied with partial rephcation to achieve load balance and high performance b this paper we amdyze the perform~ce of NO when the set, of disks used to store data blocks is not homogeneous, having both tierent bandwidths and Merent storage capacities.The basic problem to be addr~ tems, Intel Corp., md NSF -t ~-952717S.t.USO i~ith Universityof Sao Pmslq h= resear&k ptiidy SUP Fotied by a f~oIrshlF from C~q @til) Perrnksionlo ruake digitdorhardcopi= of d] orpanof Ihii urorkfor personalor classroomuse is granted withoutfee provided that copies are not made or distributed for profitor commercial advantage, andthat copiesbear!hk noticeand~hefull citationon thefirst page.To copyolhewise,to republishto postonsewers or lo redistributeIO lists, requires prior specific permission andlora fee.

We describe design and implementation of the RIO (randomized I/O) multimedia object server which manages a set of parallel disks and supports real-time throughput and statistical delay guarantees. This storage subsystem was implemented as part of a multimedia information server supporting multiple concurrent applications such as video on demand and 3D interactive virtual worlds. We discuss principal issues and innovations involved in design and implementation of RIO storage system, presenting experimental performance results measured on a prototype implementation. A multimedia data server must be ready to handle a variety of real-time object types (video, audio, 3D interactive virtual worlds, etc.) along with a nonreal-time workload component. Achieving simultaneously (1) high utilization and (2) low latency with high degree of certainty is the challenge. Our prototype system provides a statistical guarantee of quality of service. Our experimental results show that it is possible to achieve a very small probability of exceeding a deadline (less than 10−6), with relatively high disk utilization (70–99%, in terms of fraction of the maximum disk capacity), together with relatively small delay bound (on the order of 0.5–1.5 s), using contemporary disks. Results were achieved using random allocation of disk blocks and replication in conjunction with on-line load balancing. © 1998 John Wiley & Sons, Inc.

We study a multiprocessing computer system which accepts parallel programs that have a fork-join computational paradigm. The multiprocessing computer system under study is modeled as K homogeneous servers, each with an infinite capacity queue. Parallel programs arrive at the multiprocessing system according to a series-parallel phase type interarrival process with mean arrival rate of h. Upon the program arrival, it forks into K-independent tasks and each task is assigned to an unique server. Each task's service time has a k-stage Erlang distribution with mean service time of /spl lambda/. A parallel program is completed upon the completion of its last task. This kind of queuing model has no known closed form solution in the general (K/spl ges/2) case. In this paper, we show that by carefully modifying the arrival and service distributions at some imbedded points in time, we can obtain tight performance bounds. We also provide a computational efficient algorithm for obtaining upper and lower bounds on the expected response time. The methodology is flexible and allows one to trade-off the tightness of the bounds and computational cost.

The goal of the development of the Earth System Model Data Information System (ESMDIS) are to provide Earth scientists with: 1) an output management system of Earth System Model (ESM) to browse the metadata and retrieve a desired subset of ESM output; 2) an analysis system of ESM output and other related datasets; 3) an automated pipelining system for ESM data processing; 4) a visualization system; and 5) a Web based user interface to utilize the system. ESMDIS is based on DBMS centric approach, built upon the "BigSur" Earth science data schema, and developed using an object relational DBMS. We have built a prototype ESMDIS, and present the results of its development.

Lately, a need for uniform access to and integration of data stored in specialized, non-standard repositories such as GIS or multimedia storage servers has become apparent. The authors provide an overview of a heterogeneous geoscientific persistent object manager (geoPOM) developed at the UCLA Data Mining Laboratory. GeoPOM provides users with the "illusion" of a single object-oriented spatial data store even though the data is actually stored in several different spatial data repositories, thus, allowing users to define and handle spatial data in a uniform manner The geoPOM data model is based on the ODMG-93 standard for object-oriented data models, and the Open Geodata Consortium's (OGC) standardization effort for temporal-spatial object types (OGIS).

Product-form queuing network models have been widely used to model systems with shared resources such as computer systems (both centralized and distributed), communication networks, and flexible manufacturing systems. Closed multichain product-form networks are inherently more difficult to analyze than open networks, due to the effect of normalization. Results in workload characterization for closed networks in the literature are often for networks having special structures and only specific performance measures have been considered. In this article, we drive certain properties (insensitivity of conditional state probability distributions and fractional-linearity of Markov reward functions) for a broad class of closed multichain product-form networks. These properties are derived using the most basic flow balance conditions of product-form networks. Then we show how these basic properties can be applied in obtaining error bounds when similar customers are clustered together to speed up computation.

Recent technological advances have made multimedia on-demand services, such as home entertainment and home-shopping, important to the consumer market. One of the most challenging aspects of this type of service is providing access either instantaneously or within a small and reasonable latency upon request. In this paper, we discuss a novel approach, termed adaptive piggybacking, which can be used to provide on-demand or nearly-on-demand service and at the same time reduce the I/O demand on the multimedia storage server.

Recent technological advances have made multimedia on-demand servers feasible. Two challenging tasks in such systems are: a) satisfying the real-time requirement for continuous delivery of objects at specified bandwidths and b) efficiently servicing multiple clients simultaneously. To accomplish these tasks and realize economies of scale associated with servicing a large user population, the multimedia server can require a large disk subsystem. Although a single disk is fairly reliable, a large disk farm can have an unacceptably high probability of disk failure. Further, due to the real-time constraint, the reliability and availability requirements of multimedia systems are very stringent. In this paper we investigate techniques for providing a high degree of reliability and availability, at low disk storage, bandwidth, and memory costs for on-demand multimedia servers. 1 Introduction Recent technological advances in digital signal processing, data compression techniques, and high spe...

Balancing loads in a multi-server system can have a significant impact on performance. In this paper, we model such a system as a heterogeneous multi-server queueing system. We study the behavior of such a system operating under the minimum expected delay (MED) routing policy, i.e., an arriving customer is assigned to the queue which has the minimal expected value of unfinished work. This routing discipline can be viewed as a generalization of the join-the-shortest queue (SQ) discipline for homogeneous servers. There is no closed-form solution for this class of queueing problem. In this paper, we provide a methodology to compute upper and lower bounds on the mean response time of the system. This methodology allows one to tradeoff the tightness of the bounds and computational cost. Applications and numerical examples are presented which show how to use this methodology for deriving performance measures and also illustrating that the excellent accuracy of the computational algorithm which is achievable with modest computational cost.

Polling systems have long been the subject of study and are of particular interest in the analysis of high-speed communications networks. There are many options for the scheduling policies that can be used at each polling station (exhaustive, gated, customer limited, etc.). In addition, one can impose an upper bound on the total service time delivered to customers at a station per server visit. In the most common case the upper bound is a constant for each polling station, and the resulting system model is not Markovian even when service times and interarrival times are exponential. In the paper, a comprehensive solution is developed for the major scheduling policies with time limits for each polling station. The approach is based on studying the embedded Markov chain defined at the sequence of epochs when the server arrives at each polling station. The computation of transition probabilities for the embedded chain requires transient analysis of the Markov process describing the system evolution between epochs. Uniformization methods are used to develop efficient algorithms for the transition probabilities and for system performance measures. Example problems are solved using the techniques developed to illustrate the utility of the results.>

We propose a simple solution to the problem of efficient stack evaluation of LRU multiprocessor cache memories with arbitrary set-associative mapping. It is an extension of the existing stack evaluation techniques for all set-associative LRU uniprocessor caches. Special marker entries are used in the stack to represent data blocks (or lines) deleted by an invalidation-based cache coherence protocol. A method of marker-splitting is employed when a data block below a marker in the stack is accessed. Using this technique, one-pass trace evaluation of memory access trace yields hit ratios for all cache sizes and set-associative mappings of multiprocessor caches in a single pass over a memory reference trace. Simulation experiments on some multiprocessor trace data show an order-of-magnitude speed-up in simulation time using this one-pass technique.>

Multimedia information systems have emerged as an essential component of many application domains ranging from library information systems to entertainment technology. However, most implementations of these systems cannot support the continuous display of multimedia objects and suffer from frequent disruptions and delays termed hiccups. This is due to the low I/O bandwidth of the current disk technology, the high bandwidth requirement of multimedia objects, and the large size of these objects that almost always requires them to be disk resident. One approach to resolve this limitation is to decluster a multimedia object across multiple disk drives in order to employ the aggregate bandwidth of several disks to support the continuous retrieval (and display) of objects. This paper describes staggered striping as a novel technique to provide effective support for multiple users accessing the different objects in the database. Detailed simulations confirm the superiority of staggered strip...

One of the most important performance measures for computer system designers is system availability. Most often, Markov models are used in representing systems for dependability/availability analysis. Due to complex interactions between components and complex repair policies, the Markov model often has an irregular structure, and closed-form solutions are extremely difficult to obtain. Also, a realistic system model often has an unmanageably large state space and it quickly becomes impractical to even generate the entire transition rate matrix. In this paper, we present a methodology that can (i) bound the system steady state availability and at the same time, (ii) drastically reduce the state space of the model that must be solved. The bounding algorithm is iterative and generates a part of the transition matrix at each step. At each step, tighter bounds on system availability are obtained. The algorithm also allows the size of the submodel, to be solved at each step, to be chosen so as to accommodate memory limitations. This general bounding methodology provides an efficient way to evaluate dependability models with very large state spaces without ever generating the entire transition rate matrix.

Structural symmetries in stochastic well-formed colored Petri nets (SWN's) lead to behavioral symmetries that can be exploited by using the symbolic reachability graph (SRG) construction algorithm. The SRC allows one to compute an aggregated reachability graph (RG) and a "lumped" continuous time Markov chain (CTMC) that contain all the information needed to study the qualitative properties and the performance of the modeled system, respectively. Some models exhibit qualitative behavioral symmetries that are not completely reflected at the CTMC level. We call them quasi-lumpable SWN models. In these cases, exact performance indices can be obtained by avoiding the aggregation of those markings that are qualitatively, but not quantitatively, equivalent. An alternative approach consists of aggregating all the qualitatively equivalent states and computing approximated performance indices. In this paper, a technique is proposed to compute bounds on the performance of SWN models of this kind, using the results we have presented elsewhere. The technique is based on the Courtois and Semal bounded aggregation method.>

In this paper we present an algorithmic approach to bounding the mean response time of a multi-server system in which the minimum expected delay routing policy issued, i.e., an arriving job will join the queue which has the minimal expected value of unfinished work. We assume the queueing system to have K servers, each with an infinite capacity queue. The arrival process is Poisson with parameter λ, and the service time of server i is exponentially distributed with mean 1/μi, 1 ≤ i ≤ K. The computation algorithm we present allows one to tradeoff accuracy and computational cost. Upper and lower bounds on the expected response time and expected number of customers are computed; the spread between the bounds can be reduced with additional space and time complexity. Examples are presented which illustrate the excellent relative accuracy attainable with relatively little computation.

A stream processing approach for temporal query processing and optimization is discussed. Given properly sorted data, the implementation of temporal joins and semijoins as stream processors can be very effective. Also discussed are the tradeoffs among sort orders, the amount of local workspace, and multiple scans over input streams; in particular, the authors are interested in the effect of sort ordering on the local workspace. They present stream processing algorithms for various temporal joins and semijoins, and their workspace requirements for various data sort orderings. Finally, it is shown how semantic query optimization can play an important and natural role in optimization of temporal queries.>

Linearizer is one of the best known approximation algorithms for obtaining numeric solutions for closed-product-form queueing networks. In the original exposition of Linearizer, the computational cost was stated to be O(MK/sup 3/) for a model with M queues and K job classes. It is shown that with some straightforward algebraic manipulation, Linearizer can be modified to require a cost that is only O(MK/sup 2/).>

Tangram, an environment for modeling which is under development at UCLA, is discussed. One of the driving concepts behind Tangram has been the combination of large-scale data access and data reduction with a powerful programming environment. The Tangram environment is based on PROLOG, extending it with a number of features, including process management, distributed database access, and generalized stream processing. The authors describe the Tangram stream processor, the part of the Tangram environment performing query processing on large streams of data. The paradigm of transducers on streams is used throughout this system, providing a database flow computation capability.>

Many tools and techniques exist for the modeling and analysis of computer and communication systems. These tools are often complex and tailored to a narrow range of problems. The system analysis task often requires coordinated use of multiple tools and techniques which is not supported by currently available systems. The Tangram project goal is to develop an environment which makes a large set of tools and techniques readily accessible and is easily tailored to specialized applications.

Markov models are widely used for the analysis of availability of computer/communication systems. Realistic models often involve state space cardinalities that are so large that it is impractical to generate the transition rate matrix let alone solve for availability measures. Various state space reduction methods have been developed, particularly for transient analysis. In this paper we present an approximation technique for determining steady state availability. Of particular interest is that the method also provides bounds on the error. Examples are given to illustrate the method.

Markov models are widely used for the analysis of availability of computer/communication systems. Realistic models often involve state-space cardinalities that are so large that it is impractical to generate the transition-rate matrix let alone solve for availability measures. An approximation technique is presented for determining steady-state availability. Of particular interest is that the method also provides bounds on the error. Examples are given to illustrate the method.>

Obtains simple relationships among partial derivatives and moments of mean queue lengths in product form queueing network models. It is shown that the mean value analysis equations can be used to obtain recursive expressions which make it possible to easily evaluate these moments in a mean value analysis type recursion. These results are also considered important in applications such as optimization problems and sensitivity analysis. Finally, it is shown that some known mean value equations can be easily derived from the results proven.>

The past few years have witnessed an increasing number of large distributed computer system implementations based on local area networks. In these systems a number of resources (CPU's, file servers, disks, etc.) are shared among jobs originating at different sites. Evaluating the performance of such large systems typically requires the solution of a queueing network model with a large number of closed chains, which precludes the use of exact solution techniques. Therefore, it is important to develop accurate and cost-effective methods for the approximate analysis of closed queueing networks with many chains. In this paper, we present an approach based on the clustering of chains and service centers. The method is applicable to queueing networks with single server fixed rate, infimite server and multiple server service centers. We present the results obtained when the method is used to solve large queueing network models. Extensive comparison of this method to existing approximation techniques indicates that the approach has better accuracy/cost characteristics.

A locking protocol to coordinate access to a distributed database and to maintain system consistency throughout normal and abnormal conditions is presented. The proposed protocol is robust in the face of crashes of any participating site, as well as communication failures. Recovery from any number of failures during normal operation or any of the recovery stages is supported. Recovery is done in such a way that maximum forward progress is achieved by the recovery procedures. Integration of virtually any locking discipline including predicate lock methods is permitted by this protocol. The locking algorithm operates, and operates correctly, when the network is partitioned, either intentionally or by failure of communication lines. Each partition is able to continue with work local to it, and operation merges gracefully when the partitions are reconnected. A subroutine of the protocol, that assures reliable communication among sites, is shown to have better performance than two-phase commit methods. For many topologies of interest, the delay introduced by the overall protocol is not a direct function of the size of the network. The communications cost is shown to grow in a relatively slow, linear fashion with the number of sites participating in the transaction. An informal proof of the correctness of the algorithm is also presented in this paper. The algorithm has as its core a centralized locking protocol with distributed recovery procedures. A centralized controller with local appendages at each site coordinates all resource control, with requests initiated by application programs at any site. However, no site experiences undue load. Recovery is broken down into three disjoint mechanisms: for single node recovery, merge of partitions, and reconstruction of the centralized controller and tables. The disjointness of the mechanisms contributes to comprehensibility and ease of proof. The paper concludes with a proposal for an extension aimed at optimizing operation of the algorithm to adapt to highly skewed distributions of activity. The extension applies nicely to interconnected computer networks.

This paper descrbes two protocols for the detection of deadlocks in distributed data bases–a hierarchically organized one and a distributed one. A graph model which depicts the state of execution of all transactions in the system is used by both protocols. A cycle in this graph is a necessary and sufficient condition for a deadlock to exist. Nevertheless, neither protocol requires that the global graph be built and maintained in order for deadlocks to be detected. In the case of the hierarchical protocol, the communications cost can be optimized if the topology of the hierarachy is appropriately chosen.

A locking protocol to coordinate access to a distributed database and to maintain system consistency throughout normal and abnormal conditions is presented. The proposed protocol is robust in the face of crashes of any participating site, as well as communication failures. Recovery from any number of failures during normal operation or any of the recovery stages is supported. Recovery is done in such a way that maximum forward progress is achieved by the recovery procedures. Integration of virtually any locking discipline including predicate lock methods is permitted by this protocol. The locking algorithm operates, and operates correctly, when the network is partitioned, either intentionally or by failure of communication lines. Each partition is able to continue with work local to it, and operation merges gracefully when the partitions are reconnected.A subroutine of the protocol, that assures reliable communication among sites, is shown to have better performance than two-phase commit methods. For many topologies of interest, the delay introduced by the overall protocol is not a direct function of the size of the network. The communications cost is shown to grow in a relatively slow, linear fashion with the number of sites participating in the transaction. An informal proof of the correctness of the algorithm is also presented in this paper.The algorithm has as its core a centralized locking protocol with distributed recovery procedures. A centralized controller with local appendages at each site coordinates all resource control, with requests initiated by application programs at any site. However, no site experiences undue load. Recovery is broken down into three disjoint mechanisms: for single node recovery, merge of partitions, and reconstruction of the centralized controller and tables. The disjointness of the mechanisms contributes to comprehensibility and ease of proof.The paper concludes with a proposal for an extension aimed at optimizing operation of the algorithm to adapt to highly skewed distributions of activity. The extension applies nicely to interconnected computer networks.

The joint equihbrmm distribution of queue sizes in a network of queues containing N service centers and R classes of customers m derived The equilibrium state probabillUes have the general form P(S) = Cd(S) fl(xl)f2(x2) .fN(x~), where S is the state of the system, x, is the configuration of customers at the ~th service center, d(S) is a function of the state of the model, f, is a function that depends on the type of the zth service center, and C is a normalizing constant It is assumed that the eqmhbrlum probabfl~tles exmt and are unique Four types of service centers to model central processors, data channels, terminals, and routing delays are considered The queuemg dlSclphnes associated with these service centers include first-come-first-served, processor sharing, no queueing, and last-come-first-served Each customer belongs to a single class of customers while awaiting or receiving serwce at a service center, but may change classes and service centers according to fixed probabditms at the completion of a service request For open networks, state dependent arrival processes are considered Closed networks are those with no exogenous arrivals A network may be closed with respect to some classes of customers and open with respect to other classes of customers At three of the four types of serwce centers, the service times of customers are governed by probablhty dmtrlbutions hawng ratmnM Laplace transforms, different classes of customers hawng different distributions At first-come-first-served-type service centers, the service time distribution must be identical and exponentml for all classes of customers.Examples show how different classes of customers can affect models of computer systems.

In this paper we consider the application of the stack algorithm concept to a two-level paged storage hierarchy in which both levels are directly addressable by the central processor. Since both levels are directly addressable, pages need not reside in the first level of memory for a reference to be completed. The effectiveness of a page replacement algorithm in such a storage hierarchy is measured by the frequency of references to the first level of memory and the number of page promotions. It is shown that the stack algorithm concept can easily be extended to apply to two-level directly addressable memory hierarchies. A class of page replacement algorithms called stack replacement algorithms is defined. An efficient procedure exists for collecting data on the performance of stack replacement algorithms.

Scheduling algorithms for time shared computing facilities are considered in terms of a queueing theory model.The extremely useful limit of "processor sharing" is adopted, wherein the quantum of service shrinks to zero; this approach greatly simplifies the problem.A class of algorithms is studied for which the scheduling discipline may change for a given job as a function of the amount of service received by that job.These multilevel disciplines form a natural extension to many of the disciplines previously considered.The average response time for jobs conditioned on their service requirement is solved for.Explicit solutions are given for the system M/G/1 in which levels may be first come first served (FCFS), feedback (FB), or round-robin (RR) in any order.The service time distribution is restricted to be a polynomial times an exponential for the case of RR.Examples are described for which the average response time is plotted.These examples display the great versatility of the results and demonstrate the flexibility available for the intelligent design of discriminatory treatment among jobs (in favor of short jobs and against long iobs) in time shared computer systems.

Abstract We consider a model which is applicable to time‐multi‐plexed systems, such as multiplexed communication channels and time‐shared computing facilities. In this (processor‐sharing) queueing model, all jobs currently in the system share equally the processing capability of the server. In this paper, we investigate the processor‐sharing model for the case of bulk arrivals. The mean response time of the system as a function of required service time is derived. An example is given to show the effect of bulk arrivals versus single arrivals for a constant utilization.

A basic probability model that has arisen in the study of time-shared or multiprogrammed systems is the so-called processor-sharing model.Ill this model it is assumed that the processor is shared simultaneously by each unit in the system (e.g.job or program in comput er systems, or message in communication systems).In particular, if there are n units in the system, then any given unit is being processed at a rate which is (1/n)-th the rate at which it would be processed if it had the system to itself.We make the assumptions of a Poisson arrival process and exponential service times and then derive an expression for the Laplace transform of the waiting time distribution of an arriving unit conditioned on the service it requires and the number it finds in the system oil arrival.From this result we obtain the first two moments of the waiting times, the Laplace transform of the equilibrium waiting time distribution, and the first two moments of this latter distribution.The paper concludes with a discussion of the results, especially as they compare with similar results for the first-come-first-served discipline.

The use of multiprocessor systems consisting of identical and autonomous processors is a promising approach to the practical solution of problems arising in real-time applications and large compute-bound problems in general.However, finding techniques for obtaining efficient solutions to the related multiprocessor scheduling problems is a little-understood problem.The authors study the problem of scheduling a set of tasks whose operational precedence structure is representable as an acyclic directed graph.In scheduling tasks it is assumed that preemptions are allowed.The major results consist of the statement and proof of an efficient algorithm for finding the minimal-length preemptive schedule for tree-structured computations.

One of the important potentials of multiprocessor systems is the ability to speed the completion of a computation by concurrently processing independent portions of the job. In this paper we consider the static scheduling of computations for a system containing two indentical processors. The object is to complete the computation in the minimum amount of time. A computation is assumed to be specified as a partially ordered set of tasks and the execution time for each task. A solution for the two-machine case with preemptive scheduling is presented.