The Design and Implementation of Extensible Information Services

Guiyi Wei, Guangming Wang, Yao Zheng, Wei WangLecture Notes in Computer Science 2005 · DOI · OpenAlex

Abstract

In high-performance grid computing, standard middleware often fails to provide the granular resource information required for optimal scheduling across diverse applications. This paper introduces the Extensible Information Service Architecture (EISA), a three-layered framework designed to improve resource discovery and management. By supporting the dynamic deployment of custom "Information Providers," EISA allows grid environments to move beyond generic site data, enabling more precise scheduling tailored to specific application needs, such as CPU speed, storage I/O, or network bandwidth.

Introduction: The Challenge of Information Scarcity

Standard grid middleware, such as the Globus Toolkit’s MDS, typically provides a limited, default set of site information. This "one size fits all" approach creates an information scarcity that prevents schedulers from making optimal decisions for specialized tasks. For instance, a Monte-Carlo simulation requires different resource metrics than a distributed data-sharing system or a fluid dynamics model.

Manually reconfiguring remote systems to provide better data is often impractical. EISA aims to bridge this gap by creating a system where new types of resource information can be added and standardized dynamically without disrupting existing infrastructure. The system formally defines a deployment as a tuple $D < P, IN, OUT, S >$, where $P$ is the provider program, $IN/OUT$ are the parameter sets, and $S$ represents the semantic rules for transformation.

The EISA Architecture

The authors propose a three-layer architecture to manage the entire lifecycle of resource information, from raw data collection to global registration.

The high-level architecture of the EISA framework, showing the interaction between resource layers, registries, and clients.

1. Information Collection Layer

This layer serves as a container for Information Providers (IPs). It interacts directly with local resources like kernels, storage systems, and databases. It maintains an "IP pool," handles the discovery of raw data, and manages resource reservations.

2. Normalization and Organization Layer

Raw data from heterogeneous sources often arrives in incompatible formats (e.g., binary, ASCII, or comma-delimited). This layer is responsible for transforming this data into a consistent, standard format using XML naming schemas (such as the GLUE schema). This ensures the grid maintains a homogeneous view of diverse resources.

The normalization and organization process, illustrating the transformation of raw data into standardized XML formats.

3. Registration and Deployment Layer

This final layer manages interactions with the wider Virtual Organization (VO). It handles security through the Grid Security Infrastructure (GSI) and registers site capabilities with a Grid Information Index Service (GIIS), allowing clients to find and query the enhanced resources.

Key Insights and Implementation

A core insight of the EISA design is the balance between autonomy and standardization. While the grid requires standard data, individual resource providers need the autonomy to define their own discovery programs. EISA allows providers to "wrap" custom discovery logic so it appears standard to the rest of the grid. By using a dedicated normalization layer, they decouple raw data collection from the high-level metadata used by schedulers.

The system was implemented and tested within the MASSIVE (Multidisciplinary ApplicationS-oriented SImulation and Visualization Environment) project at Zhejiang University. Key results include:

  • Full Compatibility: Integration with Globus Toolkit and OGSA standards.
  • Enhanced Metadata: Successful delivery of non-default information, including network data flow and visualization device statuses.
  • Standardized Output: Compatibility with LDIF ensures that legacy discovery tools can still utilize the enhanced data.

Limitations and Future Outlook

While EISA provides a robust framework for extensibility, several areas for future work remain. The performance overhead introduced by the three-layer XML normalization process has not yet been quantitatively evaluated; in high-speed grids, translation latency could impact scheduling speed. Additionally, as the number of custom Information Providers grows, managing the IP pool without conflicts may become complex. Future research could also explore automated methods for mapping raw system metrics to high-level grid schemas, reducing the manual effort currently required to define transformation rules.

Summary updated

BibTeX
@inproceedings{wei2005design,
  author = {Guiyi Wei and Guangming Wang and Yao Zheng and Wei Wang},
  title = {{The Design and Implementation of Extensible Information Services}},
  booktitle = {Lecture Notes in Computer Science},
  year = {2005},
  doi = {10.1007/11428831\_123},
}

Loading related content...

Figure