Traffic engineering has received substantial attention in recent years as ISPs deal with exponential growth in data traffic.

Briefly, traffic engineering refers to the optimization of network resources through routing given information about the amount of traffic entering and leaving the network. Although traditionally concerned with the management of flows within a network, traffic engineering has expanded to include management of network resources through the setting of routes at the inter-domain level. The output of traffic engineering is typically a set of routes that provide the level of performance required by the ISP. Protocols such as OSPF [14], IS-IS [3], or MPLS [17] are then used to instantiate these paths, either exactly or approximately. Traffic engineering traditionally generates a set of routes between ingress and egress points in the network [8], [22], [19]. However, these are point to point routes and, thus are not always robust to failures or unpredicted changes in traffic loads. Some work has focussed on computing routes that are robust to changes in loads, [2], [25]. However, the routes are not robust to failures and there is no guarantee that they can achieve the capacity of the network. More recently there has been interest in inter-domain traffic engineering. However, this is in its infancy; see [7] for some initial results. Research supported by NSF grants CNS 05-19691, CNS 05-19922, CCF 06-34891, CNS 07-21286, and CNS 07-21779 In this paper, we propose the use of the back-pressure algorithm for the purpose of traffic engineering. First proposed in [21], this is a distributed and adaptive routing/scheduling algorithm where nodes use the queue length information of neighboring nodes to make routing decisions. Packets are adaptively routed throughout the network in response to congestion information. Thus the back-pressure algorithm is resilient to link failures and topological changes. Moreover, it has been shown to be throughput optimal, i.e., if any routing algorithm can support a set of traffic flows, then the backpressure

algorithm can as well. When deployed in a wireless network, there is a network-wide scheduling problem that requires addressing due to the contention aspect of the media. However, this problem disappears in a wireline network [1]. Finally, although first proposed to handle inelastic traffic, the back-pressure algorithm in conjunction with congestion control, can be used to provide fair resource allocation to elastic traffic [12], [5], [6], [15], [20]. Although extremely simple, through its reliance on simple rules based on local neighbor information, the back-pressure algorithm suffers one serious drawback that renders it unsuitable for adoption as a traffic engineering solution within either a large network or the Internet, namely it requires the maintenance of one queue per destination at every router. Given that the Internet includes hundreds of millions of end hosts, this precludes its deployment in an environment such as the Internet. One technique for reducing this state is to aggregate destinations according to the last hop access router that it connects to. However, at the level of the Internet, this can result in hundreds of thousands of destination queues. At the level of an ISP, this might still result in 100s of destinations. In this paper, we tackle this problem of state explosion by proposing a cluster-based back-pressure scheduling/routing algorithm. This algorithm works as follows. Routers are grouped into clusters. Associated with each cluster is one or more gateway routers that allow transit into the cluster. Each router needs only to be aware of the destinations within its cluster and the gateway routers associated with the other clusters. Routing proceeds as follows. If the packet is destined within the same cluster that it was generated, then routers use standard backpressure. However, when a packet is generated at a source destined to a different cluster, loose source routing is used.

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE INFOCOM 2008 proceedings. 978-1- robust to failures or unpredicted changes in traffic loads. Some work has focussed on computing routes that are robust to changes in loads, [2], [25]. However, the routes are not robust to failures and there is no guarantee that they can achieve the capacity of the network. More recently there has been interest in inter-domain traffic engineering. However, this is in its infancy; see [7] for some initial results. Research supported by NSF grants CNS 05-19691, CNS 05-19922, CCF 06-34891, CNS 07-21286, and CNS 07-21779 In this paper, we propose the use of the back-pressure algorithm for the purpose of traffic engineering. First proposed in [21], this is a distributed and adaptive routing/scheduling algorithm where nodes use the queue length information of neighboring nodes to make routing decisions. Packets are adaptively routed throughout the network in response to congestion information. Thus the back-pressure algorithm is resilient to link failures and topological changes. Moreover, it has been shown to be throughput optimal, i.e., if any routing algorithm can support a set of traffic flows, then the backpressure