Industrial Internet of Things (IIoT) applications can greatly benefit from leveraging edge computing. For instance, applications relying on deep neural network (DNN) models can be sliced and distributed across IIoT devices and the network edge to reduce inference latency. However, low network performance between IIoT devices and the edge often becomes a bottleneck. In this study, we propose ScissionLite, a holistic framework designed to accelerate distributed DNN inference using lightweight data compression. Our compression method features a novel lightweight down/upsampling network tailored for performance-limited IIoT devices, which is inserted at the slicing point of a DNN model to reduce outbound network traffic without causing a significant drop in accuracy. In addition, we have developed a benchmarking tool to accurately identify the optimal slicing point of the DNN for the best inference latency. ScissionLite improves inference latency by up to 15.7× with minimal accuracy degradation.

In cloud computing, a shared storage server, which provides a network-attached storage device, is usually used for centralized data management. However, when multiple virtual machines (VMs) concurrently access the storage server through the network, the performance of each VM may decrease due to limited bandwidth. To address this issue, a flash-based storage device such as a solid state drive (SSD) is often employed as a cache in the host server. This host-side flash cache saves remote data, which are frequently accessed by the VM, locally in the cache. However, frequent VM migration in the data center can weaken the effectiveness of a host-side flash cache as the migrated VM needs to warm up its flash cache again on the destination machine. This study proposes Cachemior, Firepan, and FirepanIF for rapid flash-cache migration in cloud computing. Cachemior warms up the flash cache with a data preloading approach using the shared storage server after VM migration. However, it does not achieve a satisfactory level of performance. Firepan and FirepanIF use the source node’s flash cache as the data source for flash cache warm-up. They can migrate the flash-cache more quickly than conventional methods as they can avoid storage and network congestion on the shared storage server. Firepan incurs downtime of the VM during flash cache migration for data consistency. FirepanIF minimizes the VM downtime with the invalidation filter, which traces the I/O activity of the migrated VM during flash cache migration in order to invalidate inconsistent cache blocks. We implement and evaluate the three flash cache migration techniques in a realistic virtualized environment. FirepanIF demonstrates that it can improve the performance of the I/O workload by up to 21.87% compared to conventional methods.

Convolutional neural networks (CNNs) have enabled effective object detection tasks in bioimages. Unfortunately, implementing such an object detection model can be computationally intensive, especially on resource-limited hardware in a laboratory or hospital setting. This study aims to develop a framework called BioEdge that can accelerate object detection using Scaled-YOLOv4 and YOLOv7 by leveraging edge computing for bioimage analysis. BioEdge employs a distributed inference technique with Scaled-YOLOv4 and YOLOv7 to harness the computational resources of both a local computer and an edge server, enabling rapid detection of COVID-19 abnormalities in chest radiographs. By implementing distributed inference techniques, BioEdge addresses privacy concerns that can arise when transmitting biomedical data to an edge server. Additionally, it incorporates a computationally lightweight autoencoder at the split point to reduce data transmission overhead. For evaluation, this study utilizes the COVID-19 dataset provided by the Society for Imaging Informatics in Medicine (SIIM). BioEdge is shown to improve the inference latency of Scaled-YOLOv4 and YOLOv7 by up to 6.28 times with negligible accuracy loss compared to local computer execution in our evaluation setting.

In cloud systems, computing resources, such as the CPU, memory, network, and storage devices, are virtualized and shared by multiple users. In recent decades, methods to virtualize these resources efficiently have been intensively studied. Nevertheless, the current virtualization techniques cannot achieve effective I/O virtualization when packets are transferred between a virtual machine and a host system. For example, VirtIO, which is a network device driver for KVM-based virtualization, adopts an interrupt-based packet-delivery mechanism, and incurs frequent switch overheads between the virtual machine and the host system. Therefore, VirtIO wastes valuable CPU resources and decreases network performance. To address this limitation, this paper proposes an adaptive polling-based network I/O processing technique, called NetAP, for virtualized environments. NetAP processes network requests via a periodical polling-based mechanism. For this purpose, NetAP adopts the golden-section search algorithm to determine the near-optimal polling interval for various workloads with different characteristics. We implement NetAP in a Linux kernel and evaluated it with up to six virtual machines. The evaluation results show that NetAP can improve the network performance of virtual machines by up to 31.16%, while only using 32.92% of the host CPU time used by VirtIO for packet processing.