Bundle Loss Control

The basic building block for enabling OverQoS to achieve loss control over a bundle is the Controlled-loss Virtual Link (CLVL) abstraction. The CLVL abstraction provides a bound, $q$, on the loss rate seen by the bundle over a certain period of time regardless of how the underlying network loss rate varies with time. Overlays can achieve this bound by recovering from network losses using a combination of Forward Error Correction (FEC) and packet retransmissions in the form of ARQ. By setting $q$ to an arbitrarily low value (close to 0), a CLVL provides the notion of a near-loss free pipe across a virtual link. Therefore, a CLVL isolates the losses experienced by the bundle from the loss-rate variations in the underlying IP network path. The biggest challenge in constructing a CLVL is to achieve the loss bound $q$ in the presence of time-varying cross traffic and network conditions. Additionally, the amount of bandwidth overhead should be minimized. In Section 3.2, we present a hybrid FEC/ARQ solution which minimizes the amount of redundancy required to provide a CLVL abstraction for a given value of $q$.

The total traffic between two overlay nodes consists of: (a) the traffic of the bundle; (b) the redundancy traffic required to achieve the target loss rate, $q$. The fairness and stability constraints limits the maximum rate (inclusive of the redundancy traffic) at which OverQoS can transmit across a virtual link. Let $b(t)$ denote this traffic bound at time $t$ (Section 3.1 elaborates on how $b$ is computed). Let $r(t)$ denote the fraction of redundancy traffic required by OverQoS to achieve $q$. Then, the available bandwidth for the flows in the bundle is $c(t)=b(t) \times (1-r(t))$. Thus, the service provided by a CLVL to the bundle is: As long as the arrival rate of the bundle at the entry node does not exceed $c(t)$, the packet loss rate across the virtual link will not exceed $q$, with high probability.