Evaluation of energy consumption and data access time in data fetching in grid-based data-intensive applications

Abstract

Data-intensive applications that involve large amounts of data generation, processing and transmission, have been operated with little attention to energy efficiency. Issues such as management, movement and storage of huge volumes of data may lead to high energy consumption. Replication is a useful solution to decrease data access time and improve performance in these applications, but it may also lead to increase the energy spent in storage and data transmission, by spreading large volumes of data replicas around the network. Thus, utilizing effective strategies for energy saving in these applications is a very critical issue from both the environmental and economical aspects. In this paper, at first we review the current data replication and caching approaches and energy saving methods in the context of data replication. Then, we propose a model for energy consumption during data replication and, finally, we evaluate two schemes for data fetching based on the two critical metrics in Grid environments: energy consumption and data access time. We also compare the gains based on these metrics with the no-caching scenario by using simulation.

A Survey on Modeling Energy Consumption of Cloud Applications: Deconstruction, State of the Art, and Trade-Off Debates

“…The first type focuses on data reading/writing from/to where the data are stored, while the second type emphasizes data transmission through the network. Although not specified in every study, these two element types usually coexist with each other in Cloud applications (e.g., the data fetching requires both disk reading and network transferring [60]). When it comes to Reading & Writing only, one trend is that disk IO is more powerconsuming than memory IO, while another trend is that data writing is generally more power-expensive than reading [10].…”

“…Disk Speed: Among all the indexes of a storage device, the disk speed is emphasized in the energy expenditure of an application's storage I/O operations. [60]. The power characteristics of disk speed and other indexes are essentially determined by storage device manufacturers.…”

“…2) Number of Data Sites: Spreading data across different sites is a common practice to improve data availability. Correspondingly, for a Cloud application, the more sites need to be visited, the more energy and time will be consumed for more data transmissions [60].…”

“…2) Overall Data Size: The existing studies exhibit a consensus on the positive correlation between the overall data size and the energy consumption of a Cloud application, even though the correlation was studied in various contexts. For instance, the input data size is a major driver behind the computation workload [37,39]; the energy incurred by accessing activities mainly depends on the data length [60]; and the amount of data to be transmitted is one of the discriminating factors for communication energy cost [62,45,58]. In the context of communication, the relevant studies further distinguish between two scenes: The first is on the traffic volumes exchanged between the client and the Cloud [40,1,50,51,68,66,42,75,31,81,67,43,49,29,70], while the second is on the data segments involved in, and transferred between, individual application tasks [73,57,64,41,74,2,69].…”

“…By focusing on the routers only in the network path of a Cloud application, the study [60] simplified the Internet architecture, and used the number of routers and their power profiles to model the data transmission energy:…”