Neal Sample

Scheduling under Uncertainty: Planning for the Ubiquitous Grid

Computational Grid projects are ushering in an environment where clients make use of resources and services that are far too expensive for single clients to manage or maintain. Clients compose a megaprogram with services offered by outside organizations. However, the benefits of this paradigm come with a loss of control over job execution with added uncertainty about job completion. Current techniques for scheduling distributed services do not simultaneously account for autonomous service providers whose performance, reliability, and cost are not controlled by the service user. We propose an approach to scheduling that compensates for this uncertainty. Our approach builds initial schedules based on cost estimates from service providers and during program execution monitors job progress to determine if future deadlines will be met. This approach enables early hazard detection and facilitates schedule repairs to compensate for delays.

CLAM: Composition Language for Autonomous Megamodules

Advances in computer networks that support the invocation of remote services in heterogeneous environments enable new levels of software composition. In order to manage composition at such a high level we envision a need for purely compositional languages. We introduce the CLAM composition language, a megaprogramming language. By breaking up the traditional CALL statement the CLAM language focuses on the asynchronous composition of large-scale, autonomous modules. Furthermore the language has the capability to support various optimizations that are specific to software composition.

A parallel index for semistructured data

Database systems are increasingly being used to manage semistructured data, which may not have a fixed structure or set of relationships between data items. Indexes which use tree structures to manage semistructured data become unbalanced and difficult to parallelize due to the complex nature of the data. We propose a mechanism by which an unbalanced vertical tree is managed in a balanced way by additional layers of horizontal index. Then, the vertical tree can be partitioned among parallel computing nodes in a balanced fashion. We discuss how to construct, search and update such a horizontal structure using the example of a Patricia trie index. We also present simulation results that demonstrate the speedup offered by such parallelism, for example, with three-way parallelism, our techniques can provide almost a factor of three speedup.

Extensible data management in the middle-tier

Current data management solutions are optimized for intra-enterprise, client-server applications. They depend on predictability, predefined structure and universal administrative control, and cannot easily cope with change and lack of structure. However, modern e-commerce applications are dynamic, unpredictable, organic, and decentralized, and require adaptability. eXtensible Data Management (XDM) is a new approach that enables rapid development and deployment of networked, data-intensive services by providing semantically-rich, high-performance middle-tier data management, and allows heterogeneous data from different sources to be accessed in a uniform manner. We discuss how middle tier extensible data management can benefit an enterprise, and present technical details and examples from the Index Fabric, an XDM engine we have implemented.

A comprehensive model for arbitrary result extraction

Within the realms of workflow management and grid computing, scheduling of distributed services is a central issue. Most schedulers balance time and cost to fit within a clients budget, while accepting explicit data dependencies between services as the best resolution for scheduling. Results are extracted from one service in total, and then simply forwarded to the next service. However, distributed objects and remote services adhere to various standards for data delivery and result extraction. There are multiple means of requesting results and multiple ways of delivering those results. By examining several popular and idiosyncratic methods, we have developed a comprehensive model that combines the functionality of all component models. This model for arbitrary result extraction from distributed objects provides increased flexibility for object users, and an increased audience for module providers. In turn, intelligent schedulers may leverage these result extraction features.

CPAM, A Protocol for Software Composition

Software composition is critical for building large-scale applications. In this paper, we consider the composition of components that are methods offered by heterogeneous, autonomous and distributed computational software modules made available by external sources. The objective is to compose these methods and build new applications while preserving the autonomy of the software modules. This would decrease the time and cost needed for producing and maintaining the added functionality. In the following, we describe a high-level protocol that enables software composition. CPAM, CHAIMS Protocol for Autonomous Megamodules, may be used on top of various distribution systems. It offers additional features for supporting module heterogeneity and preserving module autonomy, and also implements several optimization concepts such as cost estimation of methods and partial extraction of results.

Middle-Tier Extensible Data Management

Current data management solutions are largely optimized for intra-enterprise, client–server applications. They depend on predictability, predefined structure, and universal administrative control, and cannot easily cope with change and lack of structure. However, modern e-commerce applications are dynamic, unpredictable, organic, and decentralized, and require adaptability. eXtensible Data Management (XDM) is a new approach that enables rapid development and deployment of networked, data-intensive services by providing semantically-rich, high-performance middle-tier data management, and allows heterogeneous data from different sources to be accessed in a uniform manner. Here, we discuss how middle tier extensible data management can benefit an enterprise, and present technical details and examples from the Index Fabric, an XDM engine we have implemented.

Guidelines for safe simulation and synthesis of implicit style Verilog

We discuss the classes of machines for which implicit style design is appropriate, and give guidelines for safe simulation and synthesis of implicit style Verilog that ensure the results of cycle based simulation agree with the results of synthesis. We also propose a minor revision to IEEE 1364 for bottom testing loops that improves the clarity of safe implicit style Verilog.

Indexing Open Schemas

Significant work has been done towards achieving the goal of placing semistructured data on an equal footing with relational data. While much attention has been paid to performance issues, far less work has been done to address one of the fundamental issues of semistructured data: schema evolution. Semistructured indexing and storage solutions tend to end where schema evolution begins. In practice, a real promise of semistructured data management will be realized where schemas evolve and change. In contrast to fixed schemas, we refer to schemas that grow and change as open schemas. This paper addresses the central complications associated with indexing open and evolving schemas: we specify the features and functionality that should be supported in order to handle evolving semistructured data. Specific contributions include a map of the steps for handling open schemas and an index for open schemas.