A cost minimization data allocation algorithm for dynamic datacenter resizing

Modern datacenters dynamically adjust the number of active servers in different geographic regions to adapt to the dynamic workloads from user requests and electricity price heterogeneity. One of the main challenges for datacenter resizing is that the heavy network traffic among datacenters causes s...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Journal of parallel and distributed computing Ročník 118; s. 280 - 295
Hlavní autoři: Chen, Wuhui, Paik, Incheon, Li, Zhenni, Yen, Neil Y.
Médium: Journal Article
Jazyk:angličtina
Vydáno: Elsevier Inc 01.08.2018
Témata:
Big data processing
Cost minimization
Data allocation
Datacenter resizing
MapReduce operation
Datacenter resizing
Cost minimization
Big data processing
MapReduce operation
Data allocation
ISSN:0743-7315, 1096-0848
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Modern datacenters dynamically adjust the number of active servers in different geographic regions to adapt to the dynamic workloads from user requests and electricity price heterogeneity. One of the main challenges for datacenter resizing is that the heavy network traffic among datacenters causes significant deterioration of the overall performance and considerably increases the operational expenditure of datacenters. In this paper, we propose an efficient data allocation technique that considers both the static and dynamic characteristics of datacenters to enable more efficient datacenter resizing. We first formulate the optimal data allocation problem, propose a generic model for minimizing the communicating cost in datacenter resizing, and show that the data allocation problem is NP-hard. To produce feasible solution in polynomial time, we propose a heuristic algorithm considering the traffic flow in the network topology of datacenters by first transforming the data allocation problem into a chunk distribution tree (CDT) construction problem, and then reducing the CDT construction to a graph partitioning problem. The experimental results show that our efficient data allocation approach can improve the performance of MapReduce operations effectively with lower communicating and computing costs for datacenter resizing. •Study of the problem of optimal data allocation for datacenter resizing when adjusting the number of activated servers dynamically in geographically distributed datacenters.•Development of a heuristic algorithm to solve the optimal data allocation problem for datacenter resizing.•Evaluation of our approach on both small- and large-scale clusters.
ISSN:0743-7315
1096-0848
DOI:10.1016/j.jpdc.2017.03.010