On The Impact of Aggregation on The Performance of Traffic Aware Routing A. Sridharan S. Bhattacharyya, C. Diot R. Gu´ erin J. Jetcheva , N. Taft U. Pennsylvania Sprint Labs U. Pennsylvania Sprint Labs Philadelphia, PA Burlingame, CA Philadelphia, PA Burlingame, CA This paper investigates the impact of traffic aggregation on the performance of routing al- gorithms that incorporate traffic information. We focus on two issues. Firstly, we explore the relationship between average network performance and the coarseness (granularity) of traffic splitting across routes. Specifically, we are interested in how average network performance im- proves with our ability to distribute traffic arbitrarily across multiple paths. Secondly, we shift our attention from average to short-term performance, with again a focus on the impact of traf- fic granularity. In particular, we explore the relation between the level of traffic aggregation and its variability, which directly affects short-term routing performance. Our investigation relies on traffic traces collected from an operational network, and its results provide insight into the cost-performance trade-off associated with deploying “traffic aware” routing protocols. 1. Introduction As IP networks become the life-line of business and commercial applications, the need for better service guarantees and improved performance are driving the deployment of service dif- ferentiation and traffic engineering in IP networks. Both typically involve data path mecha- nisms like packet classification etc. and control path mechanisms like signalling and extensions to routing protocols. In this paper we focus on routing, in particular, on evaluating the trade-off that exists between the added complexity and cost of the extensions required to accommodate traffic engineering, and the performance benefits it affords. We believe that such an understand- ing is important to decide whether or not traffic aware routing is worth deploying. Traffic aware routing consists of protocols and algorithms that incorporate in the computation of routes the knowledge of both available network resources, e.g., available link bandwidth, and traffic requirements. The goal is some optimization of network usage or service guarantees. There have been many studies devoted to the design and evaluation of traffic aware routing algorithms and protocols, and they can be broadly classified in two categories. Those with a traffic engineering focus, and those that target an on-demand model (see [5,9,2] for examples of the first, and [4,1] and [8] despite its title for examples of the second). The focus of this paper is on the traffic engineering usage of routing, where a traffic matrix characterizing the bandwidth Authors emails: ashwin@ee.upenn.edu, (supratik,cdiot)@sprintlabs.com, guerin@ee.upenn.edu, jorjeta@cs.cmu.edu, nina@sprintlabs.com The work of these authors was supported in part by a grant from Sprint Labs and by NSF grants ANI99-06855 and ANI99-02943 This author is currently affiliated with CMU, but this work was done during an internship at Sprint Labs.