STLGT: A Scalable Trace-Based Linear Graph Transformer for Tail Latency Prediction in Microservices

LLM serving exhibits extreme length variability, making size-based scheduling difficult in practice. Recent LLM schedulers approximate SJF/SRPT using predicted decode lengths or ranks and primarily report mean-centric metrics such as TTFT and TBT. We show that these prediction-driven policies can be fragile under distribution shifts, bursty arrivals, and GPU memory pressure, while offering limited control over the tail latency (P90-P99) that dominates user experience, even with perfect decode-length knowledge. We introduce a distribution-aware, prediction-free scheduling framework that replaces explicit length prediction with soft priority boosting driven by lightweight statistical signals. Our design co-optimizes scheduling and cache-aware preemption to account for memory-coupled decode dynamics across workload mixes. Evaluated on production and open-source traces, our method reduces P99 TTLT by up to 35-50% relative to SRPT with perfect length knowledge and reduces TTFT by 34-47% across workloads, including reasoning-heavy and chat-heavy tasks. These results demonstrate a robust alternative for optimizing tail latency in online LLM serving.

Accurate cloud workload forecasting is pivotal for efficient resource management but remains challenging as workloads are highly volatile and prone to sudden bursts. Although wavelets preserve temporal locality, rigid fixed bases struggle with complex patterns and isolated processing neglects critical spatial dependencies. To address this, we propose SWIFT, a pure convolutional framework designed for high-efficiency workload forecasting. We introduce a Learnable Cascaded Wavelet Path that reformulates the traditional fixed wavelet bases into adaptive convolutional operators, enabling precise, data-driven feature peeling. Complementing this, our Multivariate Interaction Module sequentially models inter-variable spatial and intra-variable feature interactions to stabilize and refine noisy workload states. Extensive experiments demonstrate that SWIFT achieves SOTA accuracy with linear O(L) complexity, reducing prediction error by up to 31.04% while cutting latency by 79.74%.

The surge of large language model (LLM) applications on personal devices imposes massive, bursty workloads on cloud serving infrastructure. While prefill-decode disaggregation improves throughput and scalability, memory-bound decode instances often suffer from persistent load imbalance, as output lengths are unknown when requests arrive at the cloud. To address this, we propose MAPS, a Memory-Aware Predictive Scheduling framework tailored for disaggregated LLM serving. MAPS performs device-assisted speculative output length prediction overlapped with cloud-side prefilling, incurring negligible latency overhead. To handle generation uncertainty, MAPS applies uncertainty-aware calibration to derive output-length upper bounds with target coverage, enabling safe scheduling decisions. Building on these bounds, MAPS employs a hierarchical global-local scheduling strategy to mitigate inter-decoder queue buildup and intra-decoder head-of-line blocking. Extensive experiments on two real-world workloads and two LLMs show that MAPS significantly outperforms three state-of-the-art systems, reducing average end-to-end latency by 42.6 and tail latency by up to 84.8.