Placing Relays for Path Diversity


Relay Placement Problem

To increase reliability and robustness of mission-critical services in the event of network failures, it is often desirable and beneficial to take advantage of path diversity provided by the network topology. One way of achieving this inside a single Autonomous System (AS) is to use two paths between every Origin-Destination (OD) pair. The first path is defined by the intra-domain routing protocol running within the AS; the second path is defined as an overlay path that passes through a strategically placed relay node inside the AS. The key question then is how to place such relay nodes inside an AS, which is the focus of our work.

Methodology

We take a graph-theoretic approach in viewing the network and propose our algorithms for relay placement. Even though completely disjoint paths are preferred, we note that it is often not possible to find such overlay paths in real networks. Towards that end, we and formalize the intuitive notion of penalty for partially disjoint paths and calculate the the fraction of traffic that goes via the overlapped links as the penalty of using partially disjoint paths. The metric turns out to be very useful in measuring how two paths are disjoint in general. We propose two heuristic algorithms for relay placement: Greedy and Local.

Key Results

We apply our algorithms on three different types of topology data - real, inferred, and synthetic - and demonstrate that our method increases networkwide resilience against a single point of failure. Using daily topology snapshots and network events log obtained from an operational tier-1 ISP and hypothetical traffic matrix, we obtained the following results.
  • Relay nodes selected by our algorithm provide complete protection against 75.3% of failure events and over 99% protection against 92.8% of failure events.
  • Relay placement turn out to be not so sensitive to the topology dynamics - we note that a small number of relays selected based on network snapshot provided good quality path diversity for over six months in the ISP network.
  • Our algorithms are heuristic, however they perform close-to-optimal.

    On Going

    We are searching for a number of ways to extend our current work.
  • Incorporate layer-2 link topology for more robust path diversity.
  • Real implementation of relays within an ISP network.
  • Interdomain relay placement problem.

    Papers

    Presentation

    People