Virginia Mary Lo

Distributed shared memory: a survey of issues and algorithms

An overview of distributed shared memory (DSM) issues is presented. Memory coherence, design choices, and implementation methods are included. The discussion of design choices covers structure and granularity, coherence semantics, scalability, and heterogeneity. Implementation issues concern data location and access, the coherence protocol, replacement strategy, and thrashing. Algorithms that support process synchronization and memory management are discussed.<<ETX>>

Heuristic algorithms for task assignment in distributed systems

Investigate the problem of static task assignment in distributed computing systems, i.e. given a set of k communicating tasks to be executed on a distributed system of n processors, to which processor should each task be assigned? The author proposes a family of heuristic algorithms for Stones classic model of communicating tasks whose goal is the minimization of the total execution and communication costs incurred by an assignment. In addition, she augments this model to include interference costs which reflect the degree of incompatibility between two tasks. Whereas high communication costs serve as a force of attraction between tasks, causing them to be assigned to the same processor, interference costs serve as a force of repulsion between tasks, causing them to be distributed over many processors. The inclusion of interference costs in the model yields assignments with greater concurrency, thus overcoming the tendency of Stones model to assign all tasks to one or a few processors. Simulation results show that the algorithms perform well and in particular, that the highly efficient Simple Greedy Algorithm performs almost as well as more complex heuristic algorithms. >

Noncontiguous processor allocation algorithms for mesh-connected multicomputers

Current processor allocation techniques for highly parallel systems are typically restricted to contiguous allocation strategies for which performance suffers significantly due to the inherent problem of fragmentation. As a result, message-passing systems have yet to achieve the high utilization levels exhibited by traditional vector supercomputers. We are investigating processor allocation algorithms which lift the restriction on contiguity of processors in order to address the problem of fragmentation. Three noncontiguous processor allocation strategies-paging allocation, random allocation, and the Multiple Buddy Strategy (MBS)-are proposed and studied in this paper. Simulations compare the performance of the noncontiguous strategies with that of several well-known contiguous algorithms. We show that noncontiguous allocation algorithms perform better overall than the contiguous ones, even when message-passing contention is considered. We also present the results of experiments on an Intel Paragon XP/S-15 with 208 nodes that show noncontiguous allocation is feasible with current technologies.

Low latency and cheat-proof event ordering for peer-to-peer games

We are developing a distributed architecture for massively-multiplayer games. In this paper, we focus on designing a low-latency event ordering protocol, called NEO, for this architecture. Previous event ordering protocols prevent several types of cheats at the expense of operating at the latency of the slowest player. We broaden the definition of cheating to include four common protocol level cheats and demonstrate how NEO prevents these cheats. At the same time, NEO has a playout latency independent of network conditions and adapts to network congestion to optimize performance.

Scalable supernode selection in peer-to-peer overlay networks

We define a problem called the supernode selection problem which has emerged across a variety of peer-to-peer applications. Supernode selection involves selection of a subset of the peers to serve a special role. The supernodes must be well-dispersed throughout the peer-to-peer overlay network, and must fulfil additional requirements such as load balance, resource needs, adaptability to churn, and heterogeneity. While similar to dominating set and p-centers problems, the supernode selection problem must meet the additional challenge of operating within a huge, unknown and dynamically changing network. We describe three generic super-node selection protocols we have developed for peer-to-peer environments: a label-based scheme for structured overlay networks, a distributed protocol for coordinate-based overlay networks, and a negotiation protocol for unstructured overlays. We believe an integrated approach to the supernode selection problem can benefit the peer-to-peer community through cross-fertilization of ideas and sharing of protocols.

Cluster computing on the fly : P2P scheduling of idle cycles in the internet

Peer-to-peer computing, the harnessing of idle compute cycles throughout the Internet, offers exciting new research challenges in the converging domains of networking and distributed computing. Our system, Cluster Computing on the Fly, seeks to harvest cycles from ordinary users in an open access, non-institutional environment. CCOF encompasses all activities involved in the management of idle cycles: overlay construction for hosts donating cycles, resource discovery within the overlay, application-based scheduling, local scheduling on the host node, and meta-level scheduling among a community of application-level schedulers. In this paper, we identify four important classes of cycle-sharing applications, each requiring its own scheduling strategy: workpile, workpile with deadlines, tree-based search, and point-of-presence. We describe a Wave Scheduler for workpile tasks that exploits idle night-time cycles using a geographic-based overlay. The scheduler incorporates a quizzing mechanism to check the correctness of results and determine trust ratings for the hosts. We also propose a Point-of-Presence Scheduler to discover and schedule hosts that meet application-specific requirements for location, topological distribution, and available resources.

A Comparative Study of Real Workload Traces and Synthetic Workload Models for Parallel Job Scheduling

Two basic approaches are taken when modeling workloads in simulation-based performance evaluation of parallel job scheduling algorithms: (1) a carefully reconstructed trace from a real supercomputer can provide a very realistic job stream, or (2) a flexible synthetic model that attempts to capture the behavior of observed workloads can be devised. Both approaches require that accurate statistical observations be made and that the researcher be aware of the applicability of a given trace for his or her experimental goals.

Cluster Computing on the Fly: resource discovery in a cycle sharing peer-to-peer system

Peer-to-peer computing, the harnessing of idle compute cycles throughout the Internet, offers exciting new research challenges in the converging domains of networking and distributed computing. Our system, Cluster Computing on the Fly, seeks to harvest cycles from ordinary users in an open access, non-institutional environment. We conduct a comprehensive study of generic searching methods in a highly dynamic peer-to-peer environment for locating idle cycles for workpile applications which are heavy consumers of cycles. We compare four scalable search methods: expanding ring, advertisement-based, random walk and rendezvous point. We model a variety of workloads, simple scheduling strategies and stabilities of hosts. Our preliminary results show that under light workloads, rendezvous point performs best, while under heavy workloads, its performance falls below the other techniques. We expected rendezvous point to consistently outperform the other search techniques because of its inherent advantage in gathering knowledge about the idle cycles. However in a peer-to-peer environment, which satisfies requests on-demand, large jobs may dominate, resulting in delays for scheduling smaller jobs.

A comparison of workload traces from two production parallel machines

The analysis of workload traces from real production parallel machines can aid a wide variety of parallel processing research, providing a realistic basis for experimentation in the management of resources over an entire workload. We analyze a five-month workload trace of an Intel Paragon machine supporting a production parallel workload at the San Diego Supercomputer Center (SDSC), comparing and contrasting it with a similar workload study of an Intel iPSC/860 machine at NASA Ames NAS. Our analysis of workload characteristics takes into account the job scheduling policies of the sites and focuses on characteristics such as job size distribution (job parallelism), resource usage, runtimes, submission patterns, and wait times. Despite fundamental differences in the two machines and their respective usage environments, we observe a number of interesting similarities with respect to job size distribution, system utilization, runtime distribution, and interarrival time distribution. We hope to gain insight into the potential use of workload traces for evaluating resource management policies at supercomputing sites and for providing both real-world job streams and accurate stochastic workload models for use in simulation analysis of resource management policies.

OREGAMI: Tools for mapping parallel computations to parallel architectures

The OREGAMI project involves the design, implementation, and testing of algorithms for mapping parallel computations to message-passing parallel architectures. OREGAMI addresses the mapping problem by exploiting regularity and by allowing the user to guide and evaluate mapping decisions made by OREGAMIs efficient combinatorial mapping algorithms. OREGAMIs approach to mapping is based on a new graph theoretic model of parallel computation called the Temporal Communication Graph. The OREGAMI software tools include three components: (1) LaRCS is a graph description language which allows the user to describe regularity in the communication topology as well as the temporal communication behavior (the pattern of message-passing over time). (2) MAPPER is our library of mapping algorithms which utilize information provided by LaRCS to perform contraction, embedding, and routing. (3) METRICS is an interactive graphics tool for display and analysis of mappings. This paper gives an overview of the OREGAMI project, the software tools, and OREGAMIs mapping algorithms.