Effective replica management for improving reliability and availability in edge-cloud computing environment
Effective replica management for improving reliability and availability in edge-cloud computing environment
复制标题
有效的副本管理可提高边缘云计算环境的可靠性和可用性
DOI：
10.1016/j.jpdc.2020.04.012
复制
发表时间：
2020-09-01
影响因子：
3.8
通讯作者：
Luo, Youlong
中科院分区：
文献类型：
--
作者：
Li, Chunlin;Song, Mingyang;Luo, Youlong
多副本策略可以为边缘云系统创建多个数据副本，并将其存储在不同的数据块中，从而提高数据可用性和数据服务质量。然而，Datallodes的存储资源有限，用户对数据的需求是时变的，不合理的数据副本数量会造成文件系统的存储负担过重或数据服务质量低下。因此，需要根据实际情况动态调整数据副本的数量。在此基础上，提出了一种基于灰色马尔可夫链的动态副本创建策略。如果需要增加副本的数量，则需要将新添加的副本放置在Datallodes上。针对副本放置过程中数据负载均衡的问题，提出了一种基于快速非支配排序遗传算法的副本放置策略。此外，针对边缘云系统中数据副本同步和失效数据块的数据恢复问题，提出了一种延迟自适应副本同步策略和一种基于负载均衡的副本恢复策略。最后通过实验验证了所提策略的有效性。(C)2020爱思唯尔公司All rights reserved.
The multi-replica strategy can create multiple data replicas for the edge cloud system and store them in different Datallodes, which improves data availability and data service quality. However, the storage resources of Datallodes are limited and the user demand for data is time-varying, the unreasonable number of data replicas will cause a high storage burden on the file system or low data service quality. Therefore, the number of data replicas needs to be dynamically adjusted according to the actual situation. Based on this, a dynamic replica creation strategy based on the gray Markov chain is proposed. If the number of replicas needs to be increased, the newly added replicas need to be placed on the Datallodes. Considering the problem of load balancing of the Datallode during replica placement, this paper proposes a replica placement strategy based on the Fast Non-dominated Sorting Genetic algorithm. In addition, considering the problem of data replica synchronization and the data recovery of failed Datallodes in the edge cloud system, this paper proposes a delay-adaptive replica synchronization strategy and a load-balancing based replica recovery strategy. Finally, the experiments prove the effectiveness of the proposed strategies. (C) 2020 Elsevier Inc. All rights reserved.