Assignment of availability objectives to components of heterogenous distributed computer systems

This paper provides a scheme to assign availability objectives to various components of a heterogenous distributed computer system using a minimum cost criteria. The availability is viewed from a user (client) perspective rather than the traditional system perspective. The user availability objectives are classified in terms of operational and inherent availabilities. Such a classification enables the designers to take into account the user expectations in a more realistic way thus minimizing the high cost of providing fault tolerant systems. The user availability objectives are then translated into the system availability objectives by relating the availability needs of each distinct user group to system components utilized by that group.     

基于Agent的复杂系统分布仿真建模方法的研究

基于Agent的分布仿真是研究大型复杂系统的一种有效的、重要的方法。为了减小复杂系统仿真的复杂度,增加仿真模型的重用和可维护性,需要研究基于Agent分布仿真的建模方法。首先对复杂系统及其特性进行了分析,对基于Agent的仿真进行了全面的论述,然后对基于Agent的复杂系统仿真中的复杂系统建模分析、Agent建模分析以及Agent的分布进行了分析,给出了基于Agent的复杂系统分布仿真的建模步骤,最后给出了在此建模思想指导下的金融证券市场的建模过程。     

Optimizing server placement in distributed systems in the presence of competition

Although the problem of data server placement in parallel and distributed systems has been studied extensively, most of the existing work assumes there is no competition between servers. Hence, their goal is to minimize read, update and storage cost. In this paper, we study the server placement problem in which a new server has to compete with existing servers for user requests. Therefore, in addition to minimizing cost, we also need to maximize the benefit of building a new server.Our major results include three parts. First, for tree-structured systems, we propose an O(|V|³k) time dynamic programming algorithm to find the optimal placement of k extra servers that maximizes the benefit in a tree with |V| nodes. We also propose an O(|V|³) time dynamic programming algorithm to find the optimal placement of extra servers that maximizes the benefit, without any constraint on the number of extra servers. Second, for general connected graphs, we prove that the server placement problems are NP-complete, and present three greedy heuristic algorithms, called Greedy Add, Greedy Remove and Greedy Add-Remove, to solve them. Third, we show that if the number of requests a server can handle (i.e., server capacity) is bounded, the server placement problem is NP-complete even for tree networks. We then derive a variation of the same set of greedy heuristic algorithms, with consideration of server capacity constraint, to solve the problem.Our experiment results demonstrate that the greedy algorithms achieve good results, when compared with the upper bounds found by a linear programming algorithm. Greedy Add performs best in the unconstrained model, yielding a benefit within 12% difference from the theoretical upper bound in average. For the constrained model, Greedy Remove performs best for smaller network sizes, while Greedy Add-Remove performs best for larger network sizes. On average, the heuristic algorithms yield a benefit within 13% difference from the theoretical upper bound in the constrained model.     

Parallelizing with BDSC,a resource-constrained scheduling algorithm for shared and distributed memory systems

We introduce a new parallelization framework for scientific computing based on BDSC, an efficient automatic scheduling algorithm for parallel programs in the presence of resource constraints on the number of processors and their local memory size. BDSC extends Yang and Gerasoulis’s Dominant Sequence Clustering (DSC) algorithm; it uses sophisticated cost models and addresses both shared and distributed parallel memory architectures. We describe BDSC, its integration within the PIPS compiler infrastructure and its application to the parallelization of four well-known scientific applications: Harris, ABF, equake and IS. Our experiments suggest that BDSC’s focus on efficient resource management leads to significant parallelization speedups on both shared and distributed memory systems, improving upon DSC results, as shown by the comparison of the sequential and parallelized versions of these four applications running on both OpenMP and MPI frameworks.     

Scheduling multiple task graphs with end-to-end deadlines in distributed real-time systems utilizing imprecise computations

In order to meet the inherent need of real-time applications for high quality results within strict timing constraints, the employment of effective scheduling techniques is crucial in distributed real-time systems. In this paper, we evaluate by simulation the performance of strategies for the dynamic scheduling of composite jobs in a homogeneous distributed real-time system. Each job that arrives in the system is a directed acyclic graph of component tasks and has an end-to-end deadline. For each scheduling policy, we provide an alternative version which allows imprecise computations, taking into account the effects of input error on the processing time of the component tasks of a job. The simulation results show that the alternative versions of the algorithms outperform their respective counterparts. To our knowledge, an imprecise computations approach for the dynamic scheduling of multiple task graphs with end-to-end deadlines and input error has never been discussed in the literature before.     

A high performance algorithm for static task scheduling in heterogeneous distributed computing systems

Effective task scheduling is essential for obtaining high performance in heterogeneous distributed computing systems (HeDCSs). However, finding an effective task schedule in HeDCSs requires the consideration of both the heterogeneity of processors and high interprocessor communication overhead, which results from non-trivial data movement between tasks scheduled on different processors. In this paper, we present a new high-performance scheduling algorithm, called the longest dynamic critical path (LDCP) algorithm, for HeDCSs with a bounded number of processors. The LDCP algorithm is a list-based scheduling algorithm that uses a new attribute to efficiently select tasks for scheduling in HeDCSs. The efficient selection of tasks enables the LDCP algorithm to generate high-quality task schedules in a heterogeneous computing environment. The performance of the LDCP algorithm is compared to two of the best existing scheduling algorithms for HeDCSs: the HEFT and DLS algorithms. The comparison study shows that the LDCP algorithm outperforms the HEFT and DLS algorithms in terms of schedule length and speedup. Moreover, the improvement in performance obtained by the LDCP algorithm over the HEFT and DLS algorithms increases as the inter-task communication cost increases. Therefore, the LDCP algorithm provides a practical solution for scheduling parallel applications with high communication costs in HeDCSs.     

Control of complex distributed systems with distributed intelligent agents

Control of spatially distributed systems is a challenging problem because of their complex nature, nonlinearity, and generally high order. The lack of accurate and computationally efficient model-based techniques for large, spatially distributed systems leads to challenges in controlling the system. Agent-based control structures provide a powerful tool to manage distributed systems by utilizing (organizing) local and global information obtained from the system. A hierarchical, agent-based system with local and global controller agents is developed to control networks of interconnected chemical reactors (CSTRs). The global controller agent dynamically updates local controller agent’s objectives as the reactor network conditions change. One challenge posed is control of the spatial distribution of autocatalytic species in a network of reactors hosting multiple species. The multi-agent control system is able to intelligently manipulate the network flow rates such that the desired spatial distribution of species is achieved. Furthermore, the robustness and flexibility of the agent-based control system is illustrated through examples of disturbance rejection and scalability with respect to the size of the network.     

Communication infrastructure in distributed scheduling

The emergence of distributed artificial intelligent (DAI) introduced a new approach to solve scheduling problems by a set of scheduling systems that interact with each other in the problem-solving process. In this paper, we describe a communication infrastructure to handle connection and communication between distributed Internet scheduling systems for distributed applications. First, we present an agent model of distributed scheduling systems where agents can communicate and coordinate activities with each other via an agent communication language. Then, we define the syntax and semantics for the agent communication languages, and negotiation mechanism. Following that, we discuss the design and development of the prototype for the multi-agent scheduling systems. We conclude with a discussion of communication issues for heterogeneous agent-based scheduling systems to solve distributed scheduling problems.     

Reliable synchronization in distributed systems

In distributed computer systems, processors often need to be synchronized to maintain correctness and consistency. Unlike shared-memory parallel systems, the lack of shared memory and a clock considerably complicates the task of synchronization in distributed systems. The objective of this article is two-fold: (1) We present a new randomized agreement algorithm to synchronize cooperating processors in a distributed system. This algorithm achieves the desired agreement in expected five rounds of message exchanges, tolerating a maximum of one-fifth of the processors failures. The algorithm belongs to the class of broadcast-based synchronization problems. (2) We present a new self-stabilization algorithm for an acyclic directed-graph structured distributed systems. This new fault-tolerant algorithm survives all imaginable faults in distributed systems. The algorithm belongs to arbiter-based and broadcast-based synchronization problems.     

Markov-chain based reliability analysis for distributed systems

In a typical distributed computing system (DCS), nodes consist of processing elements, memory units, shared resources, data files, and programs. For a distributed application, programs and data files are distributed among many processing elements that may exchange data and control information via communication link. The reliability of DCS can be expressed by the analysis of distributed program reliability (DPR) and distributed system reliability (DSR). In this paper, two reliability measures are introduced which are Markov-chain distributed program reliability (MDPR) and Markov-chain distributed system reliability (MDSR) to accurately model the reliability of DCS. A discrete time Markov chain with one absorbing state is constructed for this problem. The transition probability matrix is employed to represent the transition probability from one state to another state in a unit of time. In addition to mathematical method to evaluate the MDPR and MDSR, a simulation result is also presented to prove its correction.     

Probabilistic resource allocation in heterogeneous distributed systems with random failures

The problem of finding efficient workload distribution techniques is becoming increasingly important today for heterogeneous distributed systems where the availability of compute nodes may change spontaneously over time. Resource-allocation policies designed for such systems should maximize the performance and, at the same time, be robust against failure and recovery of compute nodes. Such a policy, based on the concepts of the Derman–Lieberman–Ross theorem, is proposed in this work, and is applied to a simulated model of a dedicated system composed of a set of heterogeneous image processing servers. Assuming that each image results in a “reward” if its processing is completed before a certain deadline, the goal for the resource allocation policy is to maximize the expected cumulative reward. An extensive analysis was done to study the performance of the proposed policy and compare it with the performance of some existing policies adapted to this environment. Our experiments conducted for various types of task-machine heterogeneity illustrate the potential of our method for solving resource allocation problems in a broad spectrum of distributed systems that experience high failure rates.     

Output stabilization of distributed bilinear systems

This paper studies regional stabilization of a distributed bilinear system evolving on a spatial domain $varOmega$. Sufficient conditions for regional weak, strong and exponential stabilization are given. Also we discuss a regional optimal stabilization problem. The obtained results are illustrated by examples and simulations.     

Spreadability and vulnerability of distributed parameter systems

In this article we focus on the mathematical development of vulnerability concept and vulnerability index and analyze the dependence of the vulnerability concept definition on that of spreadability [A. El Jai and K. Kassara, “Spreadable distributed systems”, Mathem. and Comp. model., 20, pp. 47--64, 1994; A. Bernoussi and A. El Jai, “New approach of spreadability”, Journal of Mathematical and Computer Modelling 31, pp 93--109, 2000; A. Bernoussi, A. El Jai and A. J. Pritchard, “Spreadability and evolving interfaces”, Inter. J. Syst. Sci., 32, pp. 1217--1232, 2001]. It is found that the mathematical approach of the vulnerability concept is rather attractive and can be applied to many engineering problems. To illustrate this, some examples and applications are considered.     

On principles in engineering of distributed computing systems

Engineering of distributed computing systems requires understanding of principles of complex systems, which have not been yet identified. To address the situation we use a concept of structural complexity and present results of computational experiments suggesting the possibility of a general optimality condition of complex systems. The optimality condition introduces the structural complexity of a system as a key to its optimization.     

A note on cooperating distributed grammar systems working in combined modes

We investigate the generative power of cooperating distributed grammar systems with context-free rules working in the full-competence mode in combination with another derivation mode, combined sf-mode, for short. A combined sf-mode as, for example, (sf∧?k) restricts the valid derivations such that both properties have to be satisfied. If erasing rules are allowed, then except for the (sf∧?1)-, (sf∧=1)-, and (sf∧t)-modes, it is shown that the family of recursively enumerable languages is characterized. The former two exceptions characterize the family of linear context-free languages, while the latter mode describes the family of context-free languages.     

Optimal control of distributed bilinear systems

The optimal control problem for a bilinear distributed parameter system subject to a quadratic cost functional is solved. It is shown that the optimal control is given by a convergent power series in the state with tensor coefficients.     

Scheduling in distributed systems: A cloud computing perspective

Scheduling is essentially a decision-making process that enables resource sharing among a number of activities by determining their execution order on the set of available resources. The emergence of distributed systems brought new challenges on scheduling in computer systems, including clusters, grids, and more recently clouds. On the other hand, the plethora of research makes it hard for both newcomers researchers to understand the relationship among different scheduling problems and strategies proposed in the literature, which hampers the identification of new and relevant research avenues. In this paper we introduce a classification of the scheduling problem in distributed systems by presenting a taxonomy that incorporates recent developments, especially those in cloud computing. We review the scheduling literature to corroborate the taxonomy and analyze the interest in different branches of the proposed taxonomy. Finally, we identify relevant future directions in scheduling for distributed systems.     

Beyond network simulators: Fostering novel distributed applications and protocols through extendible design

Simulation has been of paramount importance to the development of novel Internet protocols. Such an approach typically focuses on one of three domains: wireless and other link-layer technologies, routing protocols, and transport-layer mechanisms and protocols. Existing techniques can tackle well simulation at layers 2, 3 and 4 of the TCP/IP architecture, but are not flexible enough to appropriately deal with application-layer protocols. These require simulators that support the modeling of networks and components with different levels of abstraction. Simmcast is an object-oriented framework that focuses on the necessary flexibility for application-layer protocol research. A simulation can be developed by the simple extension of building blocks that closely resemble components of a real network such as hosts, links and routers. The internal complexity of these components, however, is hidden from the user, so he/she can focus on the implementation of the desired protocol characteristics. This paper describes the flexible simulation architecture proposed and instantiated through Simmcast, and draws lessons from our experience in designing, implementing and deploying it. We also present framework instances used to evaluate application-layer protocols, exemplifying how different kinds of simulations can be developed with Simmcast.     

Proactive scheduling in distributed computing—A reinforcement learning approach

In distributed computing such as grid computing, online users submit their tasks anytime and anywhere to dynamic resources. Task arrival and execution processes are stochastic. How to adapt to the consequent uncertainties, as well as scheduling overhead and response time, are the main concern in dynamic scheduling. Based on the decision theory, scheduling is formulated as a Markov decision process (MDP). To address this problem, an approach from machine learning is used to learn task arrival and execution patterns online. The proposed algorithm can automatically acquire such knowledge without any aforehand modeling, and proactively allocate tasks on account of the forthcoming tasks and their execution dynamics. Under comparison with four classic algorithms such as Min–Min, Min–Max, Suffrage, and ECT, the proposed algorithm has much less scheduling overhead. The experiments over both synthetic and practical environments reveal that the proposed algorithm outperforms other algorithms in terms of the average response time. The smaller variance of average response time further validates the robustness of our algorithm.     

Strong stable properties in distributed systems

Summary A stable property in a distributed system is a global property which once true, remains true forever. This paper refines this notion by formally introducing the concept ofstrong stable properties. A strong stable property has the nice property that it can be correctly evaluated on the consistent part of uncoordinated snapshots. Termination and deadlock are shown to be strong stable properties, whereas distributed garbage is not. We also show how to derive a simple generic algorithm for the detection of a strong stable property. The generic algorithm is illustrated by two examples: termination detection and deadlock detection. Incidentally the paper presents a very simple algorithm for termination detection. Andre Schiper has been a professor of Computer Science at EPFL (Federal Institute of Technology in Lausanne, Switzerland) since 1985, leading the Operating Systems laboratory. He graduated in Physics from the Federal Institute of technology in Zürich and received his Ph.D. in Computer Science from EPFL in 1980. In 1981–82 he spent one year at the University of Rennes, France. From 1983 to 1985, he was professor at the Engineering School in Yverdon, Switzerland. Between 1989 and 1991 André Schiper was head of the Department of Computer Science of EPFL, and during the academic year 1992–93 he was on sabbatical leave at Cornell University, Ithaca (NY). His research interests are in the areas of operating systems, distributed and fault-tolerant distributed systems, and parallelism. He is currently involved in the European Esprit project BROADCAST whose objective is the design and implementation of large scale distributed computing systems. Alain Sandoz graduated in Mathematics from the University of Neuchâtel, Switzerland, in 1984 and in Computer Science from the Federal Institute of Technology in Lausanne, Switzerland, in 1988. He received his Ph.D. in Computer Science from the Federal Institute of Technology in Lausanne in 1992. His dissertation was concerned with modelling causal relationships between transactions in distributed and replicated database systems. From 1992 to 1994 he was involved in research on fault-tolerant and large scale distributed computing systems. He is currently working on the development of information systems for the Swiss government.