30 June 2025 in ACM Transactions on Architecture and Code Optimization
Transformer-based large language model (LLM) inference serving is now the backbone of many cloud services. LLM inference consists of a prefill phase and a decode phase. However, existing LLM deployment practices often overlook the distinct characteristics of these phases, leading to significant interference. To mitigate interference, our insight is to carefully schedule and group inference requests based on their characteristics. We realize this idea in ShuffleInfer through three pillars. First, it partitions prompts into fixed-size chunks so that the accelerator always runs close to its computation-saturated limit. Second, it disaggregates prefill and decode instances so each can run independently. Finally, it uses a smart two-level scheduling algorithm augmented with predicted resource usage to avoid decode scheduling hotspots. Results show that ShuffleInfer improves time-to-first-token (TTFT), job completion time (JCT), and inference efficiency in terms of performance per dollar by a large margin, e.g., it uses 38% less resources all the while lowering average TTFT and average JCT by 97% and 47%, respectively.
19 October 2026 in ACM Transactions on Sensor Networks
Long Range Frequency Hopping Spread Spectrum (LR-FHSS) has emerged as a promising technology to extend the coverage of Low-Power Wide-Area Networks (LPWANs). However, in complex indoor environments (e.g., multi-floor buildings), its performance remains unstable due to dynamic channel conditions. Existing solutions typically adopt blacklist-based strategies that disable poor channels to enhance reliability. While effective in reducing errors, such approaches significantly degrade network capacity under high concurrency. We propose DFH , a novel frequency hopping scheme that improves reliability without discarding poor channels. The key insight is that, under a fixed channel set, different frequency hopping orders can lead to distinct transmission performance due to the interleaving mechanism in LR-FHSS. By optimizing the channel usage order rather than eliminating channels, DFH enhances reliability while preserving capacity. Specifically, we introduce hop-level Signal-to-Noise Ratio (SNR) as a fine-grained channel quality metric and build a predictive model linking hopping order to Packet Reception Rate (PRR). Based on this model, we design an interleaving-aware search algorithm to efficiently identify the optimal hopping sequence. To address multi-user collisions, we further propose a greedy frequency hopping sequence selection strategy. Extensive experiments in real network demonstrate that DFH improves PRR by up to 2.76× compared to standard LR-FHSS.