Publications

Performance Characterization of Containerized DNN Training and Inference on Edge Accelerators

Abstract: Edge devices have typically been used for DNN inferencing. The increase in the compute power of accelerated edges is leading to their use in DNN training also. As privacy becomes a concern on multi-tenant edge devices, Docker containers provide a lightweight virtualization mechanism to sandbox models. But their overheads for edge devices are not yet explored. In this work, we study the impact of containerized DNN inference and training workloads on an NVIDIA AGX Orin edge device and contrast it against bare-metal execution on running time, CPU, GPU and memory utilization, and energy consumption. Our analysis provides several interesting insights on these overheads.

Towards Efficient Scheduling of Concurrent DNN Training and Inferencing on Accelerated Edges

Abstract: Edge devices are typically used to perform lowlatency DNN inferencing close to the data source. However,with accelerated edge devices and privacy-oriented paradigms like Federated Learning, we can increasingly use them for DNN training too. This can require both training and inference workloads to be run concurrently on an edge device, without compromising on the inference latency. Here, we explore such concurrent scheduling on edge devices, and provide initial results demonstrating the interaction of training and inferencing on latency and throughput. –>