An Investigation of Inter-Domain Control Aggregation Procedures Rute Sofia 1,2 Roch Guérin 1 Pedro Veiga 2 1 ESE, University of Pennsylvania rsofia@seas.upenn.edu,guerin@ee.upenn.edu 2 DI, University of Lisbon {rute,pmv@di.fc.ul.pt} Abstract Current Quality of Service models such as those embod- ied in the Differentiated Services proposal, rely on data path aggregation to achieve scalability. Data path aggregation bundles into a single aggregate multiple flows with the same quality requirements, hence decreasing the amount of state to be kept. A similar scalability concern exists on the con- trol path, where the state required to account for individual reservations needs to be minimized. There have been sev- eral proposals aimed at control path aggregation, and the goal of this paper is to expand on these works in an attempt to gain a better understanding of the various parameters that influence the efficiency of different approaches. In par- ticular, we focus on inter-domain control aggregation, and compare an Autonomous System (AS) sink-tree based ap- proach with two examples of a shared AS segment based approach, in terms of the amount of state kept, both per AS and per edge router. Our main contributions are in pro- viding a greater understanding into the design of efficient control path aggregation methods. 1. Introduction Data path Quality of Service (QoS) issues are by now reasonably well understood, and a number of different al- ternatives have been proposed and investigated, e.g., In- tegrated Services (IntServ) [3] and Differentiated Services (DiffServ) [12], each representing a different trade-off in terms of capability and scalability. However, the same understanding is not really available for control path is- sues. The control path consists primarily of mechanisms for reserving and maintaining the necessary data path re- sources, as embodied in proposals such as Internet Stream- ing Protocol (ST-II ) [5] and Resource Reservation Proto- col (RSVP) [4], with the latter being the current solution of This work was supported by the program Programa Operacional So- ciedade de Informação (POSI), of the Portuguese Fundação para a Ciên- cia e Tecnologia and of the European Union FEDER. Scholarship refer- ence: PraxisXXI/BD/18246/98, and by NSF grants ANI-9902943, ANI- 9906855, and ITR-0085930. choice for most new IP services. The main concern of these proposals is their scalability, specially in terms of inter- domain links, that are expected to carry a large volume of individual reservation requests. Our main motivation is, therefore, to gain a better per- spective into the scalability of various control mechanisms, and their ability to handle large reservation volumes. We focus on inter-domain control reservations, as we expect them to be the most stressful in terms of scalability. Our ap- proach is not so much to propose a specific mechanism, but instead to try to gain a basic understanding of factors and parameters that affect the scalability of inter-domain control reservation mechanisms. In particular, we focus on evaluat- ing various aggregation techniques that attempt to minimize state and processing due to resource reservation on links connecting different routing domains or Autonomous Sys- tems (AS’s), i.e., inter-domain links. Information provided by the Border Gateway Protocol (BGP) [17], the current dy- namic solution for inter-domain information exchange, al- lows the use of different criteria to decide how, when, and where to aggregate reservation requests. Aggregation can, for example, be done on the basis of a single shared AS hop, or on the basis of a shared AS path segment, or simply be based on having the same destination AS, as proposed by Pan et al. [6]. These different options translate into main- taining state at different locations in the network. In gen- eral, state needs to be kept for each individual reservation at all aggregation and deaggregation points, while state is kept for aggregate reservations at all the intermediate inter- domain links they traverse. Hence, the goal of a scalable solution is to minimize the overall reservation state in the network, as well as the state that any router needs to main- tain. In addition to the state needed, a scalable solution should also take into account processing and signaling require- ments, ensuring that both are kept as low as possible. A related factor is the bandwidth efficiency of a solution, and in particular how often the bandwidth allocated to an aggre- gate reservation is updated. Ideally, updating bandwidth af- ter every change to the individual reservations of an aggre- gate would ensure that only the minimum possible quantity 1