High Performance Multimodal Networks

Networks often form the core of many users' spatial databases. Networks are used to support the rapid navigation and analysis of linearly connected data such as that found in transportation networks. Common types of analysis performed on such networks include shortest path, traveling salesman, allocation, and distance matrix computation. Network data models are usually represented as a small collection of tables: a junction table and an edge table. In the context of networks used to model transportation infrastructure, it is also necessary to model turn restrictions and impedances (delays). Network data is frequently persisted in normalized relational tables that are accessible via standard SQL-based queries. We propose a different approach where the network connectivity information is persisted using a compressed binary storage representation in a relational database. The connectivity information is accessible via standard Java, .NET, and COM APIs that are tailored to common access patterns used in the support of high performance network engines. These network engines run on the client or application server tier rather than as extensions on the relational server. In this paper, we discuss the problem of building a robust and scalable implementation of a network data model. The fundamental and central requirements are enumerated. These requirements include support for hierarchical networks, turn restrictions, and logical z elevations. We propose a different approach to representing network topology that addresses many of the high-end modeling requirements of network systems. Our approach supports all of the listed requirements in addition to multimodal modeling (e.g., coexistent road, bus, and rail networks) within the context of multi-user, long transaction databases.