Towards Delay Assurances in FIFO Networks

The goal of this paper is to examine the gains of partial upgrades to existing FIFO networks, to support delay assurances. Specifically, we try to find the number of hops of FIFO multiplexing after which a latency target is violated. We first examine the effect of multiplexing two flows through successive FIFO schedulers, and for a simple scenario where cross-traffic is assumed to be absent, we derive a worst-case bound on the burstiness increase across n nodes. We use the result to obtain an effective service curve and a worst-case latency bound. We then examine the effect of having a priority scheduler at the entry of the network with FIFO nodes in the core. We provide a basis for determining the number of hops up to which a worst-case latency target is met.