In-network aggregation is a fundamental operation for massive packets in the Internet of Things (IoT). By aggregating ultra-small packets, the energy consumption for data transmission is not related to the packet number, while the average delay performance depends on the delay of all packets even if they are aggregated. In this paper, we propose a low-delay ultra-small packet transmission scheme with in-network aggregation in energy-harvesting multi-hop networks, where each device periodically transmits packets in a collect-wait-forward relaying manner. Considering the resulting extra waiting time during relaying, we first drive the tractable form of the average end-to-end delay by problem transformation. By characterizing the two-dimensional evolution property from the perspective of both hops and time, the delay minimization problem is reformulated as an infinite-horizon average-cost Markov decision process with a two-dimensional optimality equation. To deal with the curse of dimensionality, we decompose the global Bellman equation into several per-device local relay selection problems. Based on the problem decomposition, we propose a distributed ultra-Small Packet Aggregation Relay SElection (SPARSE) algorithm via stochastic learning. The convergence is further proved theoretically and verified by simulation. Simulation results reveal that the proposed scheme achieves significant performance gain over the baselines for ultra-small packets.

Instant Messaging (IM) Service is an integrated service composed by multiple types of subservices, which have diverse performance requirements and evaluation criteria. IM service has drawn a lot of attention, yet a suitable Quality of Experience (QoE) model for IM service can rarely be found. In this paper, we propose a Representative Service (RS) based QoE modeling criterion for IM service. We discover that the QoE of IM service users mainly depends on the Representative Service Quality (RSQ) by exploiting the patterns of the IM service user behaviors. In the proposed RS based QoE model, the Normalized Quality (NQ) is defined for different subservices with various performance evaluation criteria to unify their qualities, and the Attention Factor (AF) is proposed to estimate the subservice that a user is focusing on. By performing a small-scale but illustrative subjective test, we verify the proposed QoE model and indicate the effectiveness of the model in tracing the IM service user behaviors.