Cost- and Quality-of-Service-Aware Network-Service Deployment Robert Haas , Patrick Droz and Burkhard Stiller IBM Research, Zurich Research Laboratory Computer Engineering and Networks Laboratory CH-8803 R¨ uschlikon, Switzerland ETHZ, CH-8092 Zurich, Switzerland rha,dro @zurich.ibm.com stiller@tik.ee.ethz.ch Abstract: This paper presents the information ag- gregation methods enabling cost- and QoS-aware service deployment. It is expected that network management tools will require the network itself to participate in the service deployment task so as to adapt to heterogeneous network- nodes cost, QoS, and capabilities. The aggregation meth- ods are part of a set of hierarchically-distributed compu- tations, for which a formal description is presented. Four types of information that can be handled by the mech- anism are introduced and illustrated by examples from three service-deployment categories. Keywords: Cost and quality of service awareness, infor- mation aggregation, automated service deployment 1. Introduction The importance of fast, optimized, and reliable deploy- ment of new services into a network is a key issue for Internet Service Providers. Simultaneously, network el- ements in general offer an increasing spectrum of capa- bilities, some with dedicated specialized functions, oth- ers with programmable behaviors, in soft- or hardware. These two driving forces render the deployment of ser- vices in a network more complex to manage. As a first step to enable interoperability and faster creation of new services, standardized Application Programming Inter- faces (APIs) for network elements are being defined. But in such an environment of networks with large numbers of nodes that need to be enabled with new services and that have widely varying capabilities and resources, it is necessary to define and provide a way to organize the de- ployment of new services at a network level. Here, we present the information aggregation meth- ods of a framework used to capture the dynamic capa- bilities of a network and to organize its deployment based on specific service-allocation policies, thereby addressing the programmability and heterogeneity aspects of the net- work. To our knowledge, this framework is the first to rec- ognize and propose solutions to this particular problem. The motivation of our approach can be viewed as simi- lar to that of quality-of-service (QoS)-aware routing pro- tocols. Such protocols replace lengthy and error-prone manual steps of provisioning resources within a network to guarantee a certain level of QoS. By handling the in- formation related to the available QoS in the network in- ternally, routing protocols can perform this task more ef- ficiently than any operator or management platform alone could. In addition, the cost of deploying a service can be computed in a distributed fashion as a function of perfor- mance, thereby allowing the most profitable option to be chosen. This paper is structured as follows: Section 2 reviews related activities. Section 3 introduces a formalism for hierarchically-distributed computations, illustrated with an example. Section 4 presents the various basic informa- tion types, and introduces aggregation methods for ser- vices of the three service-deployment categories. Section 5 contains a brief summary of this contribution as well as an outlook. 2. Related Work There are only few known activities that focus on the automated deployment of services over large heteroge- neous programmable networks. Hierarchical architec- tures have been used in routing protocols and network management, but not yet considered in the context of de- ploying services. Let us briefly review the main activities of related work. In [1], the need for a distributed and programmable management platform for the future Internet is presented, and simple navigation patterns for mobile agents are de- scribed. The underlying hierarchical structure in our con- tribution can be viewed as another, more sophisticated navigation pattern used by mobile agents to perform the specific task of scalable service deployment. In [2], the need for an automated design, creation, and deployment of network architectures is presented, and a