Content name resolution service implementation for cache and forward network architecture

Cache aNd Forward (CNF) is a proposed architecture for content delivery services in the future Internet. The CNF architecture takes advantage of reductions in storage to design a network that directly addresses the mobile content delivery problem. The CNF architecture uses a content name resolution service protocol, along with a reliable hop-by-hop transport protocol, storage aware routing protocol in place of end-to-end TCP for reliable delivery of large files. This thesis presents the algorithms proposed for a distributed name resolution protocol and design and experimental evaluation of the protocol on ORBIT in context of a multi-hop wireless access network scenario. The protocol is designed using hashing technique such that when a host queries for a file, the name service will be triggered and will return the addresses of nodes that cache the file. Since our architecture is about caching and forwarding large content files, enabling hosts to retrieve files from the network and not necessarily from the origin server, we need to uniquely identify the files. To that effect, we propose to identify a file using a unique content identifier (CID) where CID is obtained by a one way hashing (SHA1) on the content itself. The aim here is to optimize selection of cache location and serve the host with the file from the nearest location. If the selected cache location is determined to be temporarily degraded, either due to poor channel conditions or mobility, the protocol uses multiple hash technique to provide alternate cache locations and the decision is based on the ETT metric provided by the routing protocol. The CNRS protocol over multi-hop 802.11 access networks with CNF routers has been implemented as a real-time proof-of-concept prototype on the ORBIT testbed. Baseline results for CNRS with hop-by-hop transport show that content based CNF network architecture performs better than TCP/IP stack. Using different content distributions, we have shown that multiple hashing, popularity based and location based caching provide significant gains over the baseline algorithm.