In this study, we characterize GPU failures in Delta, the current large-scale AI system with over 600 petaflops of peak compute throughput. The system comprises GPU and non-GPU nodes with modern AI accelerators, such as NVIDIA A40, A100, and H100 GPUs. The study uses two and a half years of data on GPU errors. We evaluate the resilience of GPU hardware components to determine the vulnerability of different GPU components to failure and their impact on the GPU and node availability. We measure the key propagation paths in GPU hardware, GPU interconnect (NVLink), and GPU memory. Finally, we evaluate the impact of the observed GPU errors on user jobs. Our key findings are: (i) Contrary to common beliefs, GPU memory is over 30x more reliable than GPU hardware in terms of MTBE (mean time between errors). (ii) The newly introduced GSP (GPU System Processor) is the most vulnerable GPU hardware component. (iii) NVLink errors did not always lead to user job failure, and we attribute it to the underlying error detection and retry mechanisms employed. (iv) We show multiple examples of hardware errors originating from one of the key GPU hardware components, leading to application failure. (v) We project the impact of GPU node availability on larger scales with emulation and find that significant overprovisioning between 5-20% would be necessary to handle GPU failures. If GPU availability were improved to 99.9%, the overprovisioning would be reduced by 4x.

$ $Large language models (LLMs) have become an increasingly important workload for cloud providers catering to both enterprise and consumer applications. LLM inference requests from these applications have end-to-end latency SLOs that must be adhered to in production settings. However, existing LLM serving systems focus on optimization objectives such as request serving throughput or request execution latency rather than the end-to-end latency SLOs. Achieving end-to-end SLOs for latency-sensitive requests is challenging due to head-of-line (HOL) blocking in the request queue, which results from bursty arrival rates and insufficient resources. To address the above challenge, we propose QLM, a multi-model queue management framework for LLM serving. QLM uses stochastic programming to orchestrate the actions of multiple LLM Serving Operations (LSOs) to reduce HOL blocking and maximize SLO attainment. Specifically, QLM uses the following LSOs: model swapping, request eviction, GPU-CPU state swapping, load balancing, and warm model start. Evaluation on heterogeneous GPU devices and models with real-world LLM serving dataset shows that QLM improves SLO attainment by 40-90% and throughput by 20-400% while maintaining or improving device utilization compared to other state-of-the-art LLM serving systems.

Large language models (LLMs) have been driving a new wave of interactive AI applications across numerous domains. However, efficiently serving LLM inference requests is challenging due to their unpredictable execution times originating from the autoregressive nature of generative models. Existing LLM serving systems exploit first-come-first-serve (FCFS) scheduling, suffering from head-of-line blocking issues. To address the non-deterministic nature of LLMs and enable efficient interactive LLM serving, we present a speculative shortest-job-first (SSJF) scheduler that uses a light proxy model to predict LLM output sequence lengths. Our open-source SSJF implementation does not require changes to memory management or batching strategies. Evaluations on real-world datasets and production workload traces show that SSJF reduces average job completion times by 30.5-39.6% and increases throughput by 2.2-3.6x compared to FCFS schedulers, across no batching, dynamic batching, and continuous batching settings.