Creating a Distributed System Middle-ware: Implementing and Optimizing X-Searches 3.7

Overview

This technical report discusses the implementation of a multi-node architecture that facilitates X-Search and provides inter-node communication for transferring complex messages in a distributed system. The middleware aims to enable efficient information retrieval in parallel and distributed filesystems.

Key Components

• Leader-worker model architecture
• TCP/IP-based socket communication
• Single-threaded non-blocking I/O using select()
• Complex message handling using:
  • Google Protocol Buffer
  • Network File System (NFS)
• X-Search integration with Borus library

Implementation Details

• Leader node acts as moderator and handles connections from worker nodes
• Workers perform local indexing and searching
• File data stored in hash tables
• Tokenization and indexing process for search functionality
• Support for single term search queries

Experimental Analysis

• Tests conducted on Mystic testbed using 5 torus nodes
• Performance evaluated across:
  • Varying number of indexing threads
  • Different node configurations
  • Search latency using Protocol Buffer vs NFS
• Results showed millisecond-level search latencies across 32GB dataset

Future Work

• Implement multi-threaded non-blocking I/O
• Add support for multi-term queries
• Enable multiple results per node
• Scale experiments to more nodes
• Optimize communication model

References

The paper includes references to:

• Select() API documentation
• Google Protocol Buffer documentation
• Elasticsearch architecture
• Related work on scalable I/O and distributed systems

This research was conducted using the NSF-funded Mystic testbed.