| Abstrakt: |
As the application complexity increases, it is desired to offload the computationally intensive tasks from the end devices to the cloud. In the cloud, virtual machines (VM) conduct the computations on behalf of the end devices. To reduce the processing delay of the offloaded tasks at the cloud, VMs may be created in advance by anticipating the arrival of offload requests, which is termed 'VM pre-provisioning'. For certain types of applications, however, long delay or high cost for the communication between the end devices and the cloud is not acceptable, and edge computing is used as an alternative. Since the edge servers are typically equipped with far smaller resource capacity than the cloud servers, the conventional VM pre-provisioning becomes inefficient. In this paper, we focus on VM pre-provisioning for resource-constrained edge servers. The VMs to be pre-provisioned are carefully chosen to effectively utilize the limited resources of the edge servers. When there exist pre-provisioned VMs that match the offload requests, those VMs are cloned to serve the offload requests, which enables quick start of the VMs. Extensive evaluation shows that the proposed scheme outperforms the existing schemes in the resource-constrained edge computing environment, particularly when the application diversity is high. For realistic evaluation, we collect a large-scale measurement trace of global users and use it in conjunction with the existing Google cluster trace. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink