25 May 2020 in IEEE Transactions on Cloud Computing
Mobile applications can leverage high-quality deep learning models such as convolutional neural networks and deep neural networks to provide high-performance cognitive services. Prior work on deep learning models-based mobile applications in a cloud-edge computing environment focuses on performing lightweight data pre-processing tasks on edge servers for cloud-hosted cognitive servers. These approaches have two major limitations. First, it is uneasy for the mobile applications to assure satisfactory user experience in terms of network communication delay, because the intermediary edge servers are used only to pre-process data (e.g., images and videos) and the cloud servers are used to complete the tasks. Second, these approaches assume the pre-trained deep learning models deployed on cloud servers are static, and will not attempt to automatically upgrade in a context-aware manner. In this article, we propose a cloud-edge collaboration framework that facilitates delivering cognitive services with long-lasting, fast response, and high accuracy properties. We fist deploy a shallow model (i.e., EdgeCNN) on the edge server and a deep model (i.e., CloudCNN) on the cloud server. EdgeCNN can provide durable and rapid response cognitive services, because edge servers not only provide computing resources for mobile applications, but also close to users. Then, we enable CloudCNN to assist in training EdgeCNN to improve the performance of the latter. Thus, EdgeCNN also provides high-accuracy cognitive services. Furthermore, because users may continue to upload data to edge servers in real-world scenarios, we propose to use the ongoing assistance of CloudCNN to further improve the accuracy of the shallow model. Experimental results show that EdgeCNN can reduce the average response time of cognitive services by up to 55.08 percent and improve accuracy by up to 26.70 percent.
17 June 2021 in IEEE Access
Virtual block devices are heavily used to fulfill the block storage needs of hypervisor-based virtual machine (VM) instances through either local or remote storage spaces. However, a high degree of VM co-location makes it increasingly difficult to physically provision all the necessary block devices using only local storage space. Also, the local storage performance degrades rapidly as workloads interleave. On the other hand, when block devices are acquired through remote storage services, the aggregated network traffic may consume too much cluster-wide network bandwidth in a cloud data center. In order to solve these challenges, we propose a caching scheme for virtual block devices within the hypervisor. The scheme utilizes the physical node's finite local storage space as a block-level cache for the remote storage blocks to reduce the network traffic bound to the storage servers. This allows hypervisor-based compute nodes to serve the hosted VMs' I/O (Input/Output) requests from its local storage as much as possible while enabling VMs to exercise large storage space beyond the capacity of local disks for new virtual disks. Caching virtual disks at block-level in a cloud data center poses several challenges in maintaining high performance while adhering to the virtual disk semantics. We have realized the proposed scheme, called vStore, on Xen hypervisor nodes with factual assessment on its design effectiveness and implementation efficiency. Our comprehensive experimental evaluations show that the proposed scheme substantially reduces the network traffic (49% on average), and incurs less than 12% overheads on the storage I/O performance.