BANDS architecture VPN

In an inter-domain network, it is not scalable for a pure centralized management system to collect all the requirements in each domain network and then compute the correct policies accordingly. Therefore, we proposed ahybrid framework using a centralized and a distributed system [10]. In the BANDS architecture, we introduced arequirement server (RS) to each domain (a.k.a. Autonomous System) in the network. The requirement server isresponsible for cooperation and policy negotiation with other requirement servers at other domains. As illustratedin Fig. 8, requirement servers execute a two-phase policy negotiation process.(i) Route path discoveryIn order to discover the requirement servers along the route path and to involve them in Phase II for thepolicy negotiation, it is necessary to find out the route path first given a start and end node. Based onthe discovered AS route path, each RS in the AS on the path should be able to identify the IP addressesof other servers along the path by using DNS. This phase is to prepare for Phase II for the correspondingrequirement servers to exchange information for policy generation. For instance, if BGRP6[7] is used underBANDS architecture, the route path discovery starts with a "BGRP PROBE" message from the initiator tothe destination. After a "GRAFT" message is sent back, the exact AS route path has been probed andreserved.(ii) Requirement discovery and policy negotiationAfter the AS route path is probed in Phase I, each RS on the path needs to make queries to its neighbor RSto discover the corresponding requirements for that traffic flow. With the RS architecture shown in Fig. 9,the requirement server stores the local requirement information as well as the routing data and existingtunnel information in its MIBs (Management Information Base). To maintain an up-to-date copy of data, itneeds to periodically update these data from local routers.Once the server receives all the data it needs, it will invoke the policy negotiation module to calculatepolicies based on the corresponding requirements. In [10], we used the Direct approach to automaticallygenerate security policies. However, as this approach may produce extra tunnels, we will use the OrderedSplit algorithm [9] to generate a minimum set of policies.