Infrastructure support for accessing network services in dynamic network environments.

摘要

Advances in wireless networking and communication-enabled portable devices raise the prospect of a mobile user being able to interact with network-based services in a seamless, ubiquitous fashion. However, achieving this goal requires applications to cope with the heterogeneity and dynamic nature of resource availability in the network. Unfortunately, current infrastructures, which rely either on differentiated service or a close coupling between services and client applications to adapt to changing network conditions, are incapable of ensuring this. A more promising approach is to augment network paths between client applications and services with application-specific functionality. While several such path-based approaches have been proposed, current approaches lack mechanisms for (1) automatically creating effective network paths whose performance is optimized for encountered network conditions, (2) dynamically reconfiguring such paths when these conditions change, and (3) effectively managing network resources across networks.; This dissertation describes our solutions for these problems, which are built into a programmable network infrastructure called CANS (Composable Adaptive Network Services). The CANS infrastructure provides support for applications to bridge the bandwidth and resource gap between network services and clients, using network aware paths that are automatically created and can further be dynamically modified. Our solution highlights four key mechanisms: (a) a high-level integrated type-based specification of components and network resources; (b) an automatic path creation strategy; (c) system support for low-overhead path reconfiguration; and (d) distributed mechanisms for managing network resources.; This approach is evaluated by measuring performance of typical applications over a range of network and end-device characteristics, and simulating the behavior of applications on large scale network topologies. Our results validate the effectiveness of our approach, verifying that (1) path-based approaches provide better and more stable performance under most system configurations as compared to end-point or proxy-based approaches; (2) using our approach, automatically generated data paths produce considerable performance advantages; (3) our flexible path creation mechanisms can provide desirable adaptation behaviors with minimal input from applications; (4) despite their flexibility, both run-time overhead and reconfiguration time of CANS data paths are negligible for most applications.