Serverless computing offers a compelling paradigm for deploying machine learning inference workflows composed of heterogeneous CPU and GPU functions. However, existing data-passing solutions in serverless systems primarily rely on host memory for data exchange (host-centric), leading to substantial data movement and salient I/O overhead. Moreover, modern GPU communication libraries (e.g., NCCL, NVSHMEM, UCX) are ill-suited to serverless environments, suffering from redundant data copies, underutilized transfer bandwidth, and inefficient temporary GPU storage.

The prosperity of machine learning applications has promoted the rapid development of GPU architecture. It continues to integrate more CUDA Cores, larger L2 cache and memory bandwidth within SM. Moreover, the GPU integrates Tensor Core dedicated to matrix multiplication. Although studies have shown that task co-location could effectively improve system throughput, existing works only focus on resource scheduling at the SM level and cannot improve resource utilization within the SM. In this paper, we propose Aker, a static kernel fusion and scheduling approach to improve resource utilization inside the SM while ensuring the QoS (Quality-of-Service) of co-located tasks. Aker consists of a static kernel fuser, a duration predictor for fused kernels, an adaptive fused kernel selector, and an enhanced QoS-aware kernel manager. The kernel fuser enables the static and flexible fusion for a kernel pair. The kernel pair could be Tensor Core kernel and CUDA Core kernel, or computing-prefer CUDA Core kernel and memory-prefer CUDA Core kernel. After the kernel fuser provides multiple fused kernel versions for a kernel pair, the duration predictor precisely predicts the duration of the fused kernels and the adaptive fused kernel selector locates the optimal fused kernel version. Finally, the kernel manager invokes the fused kernel or the original kernel based on the QoS headroom of latency-critical tasks to improve the system throughput. Our experimental results show that Aker improves the throughput of best-effort applications compared with state-of-the-art solutions by 50.1% on average, while ensuring the QoS of latency-critical tasks.

Machine Learning (ML) workflows are increasingly deployed on serverless computing platforms to benefit from their elasticity and fine-grain pricing. Proper resource allocation is crucial to achieve fast and cost-efficient execution of serverless ML workflows (specially for hyperparameter tuning and model training). Unfortunately, existing resource allocation methods are static, treat functions equally, and rely on offline prediction, which limit their efficiency. In this paper, we introduce CE-scaling – a Cost-Efficient autoscaling framework for serverless ML work-flows. During the hyperparameter tuning, CE-scaling partitions resources across stages according to their exact usage to minimize resource waste. Moreover, it incorporates an online prediction method to dynamically adjust resources during model training. We implement and evaluate CE-scaling on AWS Lambda using various ML models. Evaluation results show that compared to state-of-the-art static resource allocation methods, CE-scaling can reduce the job completion time and the monetary cost by up to 63% and 41% for hyperparameter tuning, respectively; and by up to 58% and 38% for model training.