Data allocation method considering server performance and data access frequency with consistent hashing

Abstract

As the network scale expands, operation support system (OSS) will need to handle more data queries and much larger quantities of data. Distributed data storage is a key OSS technology that makes OSS highly scalable and provides high performance. Many data allocation methods have been proposed to achieve load-balancing of servers and a short response time. Consistent hashing is one effective method of data allocation, especially when the number of system devices frequently changes. However, existing methods still cause an unbalanced server load and an increase in response time because consistent hashing does not address server performance. Furthermore, these methods do not consider dynamic changes in data access frequency. We proposes a data allocation method using virtual nodes with consistent hashing that takes server performance and data access frequency into account. Simulation results showed that the proposed method achieves much better load-balancing of servers and a shorter response time than existing methods.

BoxBroker: A Policy-Driven Framework for Optimizing Storage Service Federation

“…Chang et al [22] use a probability-based algorithm in order to select the best coordination of services that achieved the maximum availability based on a given budget. Other studied approaches regarding data allocation are [23,24,25], focusing on load balancing in heterogeneous storage environments. In most of the mentioned related works, only one or two aspects of storage are considered.…”

“…In most of the mentioned related works, only one or two aspects of storage are considered. For example, [21] and [22] mainly focus on availability and cost, while [23,24,25] focus on capacity, access frequency, and performance.…”

“…18. In comparison to [14,21,22,23,24,25] approaches, where the selection is based on fixed metrics, our proposal can be used with any number of metrics defined by the clients. In addition, unlike these approaches our algorithm allows clients to change the importance of each metric depending on the data.…”