Research On Key Techniques In Content Router

Currently, the prosperous Internet applications bring many challenges to the IP network, to better satisfy the requirements of more efficient and convenient access to the content and service in the Internet, Named Data Networking(NDN) was proposed as a content-centric future Internet architecture. NDN has three important properties: 1) large volume of content can be cached at routers to satisfy subsequent requests; 2) route and forward based on content or service names; 3) stateful routing. Each property corresponds to a key module on the forwarding plane: Content Store(CS), Forwarding Information Base(FIB) and Pending Interest Table(PIT). This dissertation conducts studies on these three tables, and the main contributions are as follows:1. Content Store is a basic function of the content router, it differs from conventional packet buffering and requires unique storage architecture. This paper proposes the concept of storage card, and develops a distributed storage architecture based on this concept to implement the Content Store. Afterwards, this paper evaluates several management schemes under this hardware storage architecture by modeling and simulation. Evaluation results reveal that the consistent hashing scheme approximates the ideal performance at very low cost.2. The forwarding plane needs to sequentially lookup CS, PIT and FIB. Generally,CS and PIT require exact match, and FIB requires the longest prefix match. Meanwhile,the lookup is based on names rather than IP addresses, where name-based lookup is much more complicated than IP-based lookup. This makes the high-speed and low-latency forwarding a daunting task. To address this problem, we propose the concept of unified index,which integrates the indexes of the three tables into a single one, so that three table lookups can be reduced to only one! By this method, we simultaneously achieve high-speed and low-latency lookup.3. Meanwhile, the name-based FIB will have huge volume because it contains numerous name prefixes and each prefix is so much longer than an IP address. Therefore,an algorithm called CONSERT is further proposed to optimally compress the FIB. The optimality means that CONSERT computes optimal forwarding tables with the minimal number of prefixes, while keeping equivalent forwarding behavior. We formulate this algorithm and prove its optimality via the induction method. Evaluation results show that CONSERT can effectively reduce the number of prefixes, so as to degrade memory consumption.4. On the forwarding plane, other than keeping the routing states, the PIT has another function — request aggregation, which means multiple identical requests can be aggregated to only one, so that the traffic forwarded to the upstream is reduced. To explore the request aggregation ratio, as well as the expected number of PIT entries, a brand new observation on PIT is developed: PIT is a TTL(Time To Live) based cache. Based on this new observation, we provide analytical models for PIT, and derive formulas for the aforementioned two metrics. The model and simulation results are very close, revealing that the analytical model is reasonable and accurate.