استفاده از یک مدل بهینه سازی تبادل پیام (MEO) برای کاهش مصرف انرژی در سیستم های توزیع شده

The concept of optimizing energy efficiency in distributed systems has gained particular interest. Most of these efforts are focused on the core management concepts like resource discovery, scheduling and allocation without focusing on the actual communication method among system entities. Specifically, these do not consider the number of exchanged messages and the energy that they consume. In this work, we propose a model to optimize the energy efficiency of message-exchanging in distributed systems by minimizing the total number of messages when entities communicate. So we propose an efficient messaging-exchanging optimization (MEO) model that aims to minimize the sum of requests and responses as a whole rather than only the number of requests. The view is to optimize firstly the energy for communication (e.g. latency times) and secondly the overall system performance (e.g. makespan). To demonstrate the effectiveness of MEO model, the experimental analysis using the SimIC is based on a large-scale inter-cloud setting where the implemented algorithms offer optimization of various criteria including turnaround times and energy consumption rates. Results obtained are very supportive.

In recent years, users have increased the scale of the distributed systems due to the increased resource utilization. Clearly, the increased sum of messages that are exchanged between users and system entities have led to an increased energy consumption. Current solutions are focused on the optimization of the performance measures (e.g. resource discovery and scheduling) without focusing on the benefits that may derive from the introduction of a new message exchanging approach. This affects the amount of messages sent and received with regards to the energy consumption of nodes. This area emphasizes an increasing trend that shifts the focus from improving performance to optimizing the energy efficiency and performance of the system as a whole [5] and [8]. By planning an energy efficient computational model along with performance optimization we can reduce the consumption rates while at the same time increase the user satisfaction. In distributed systems, the performance includes the computational metrics (e.g. execution times) while the energy efficiency includes consumption of the interacting nodes. Our contribution is by improving the energy consumption rates based on the minimization of the sum of messages that are exchanged during the resource management phase. We define as message exchange in distributed systems the communication among entities (nodes). To achieve this, we introduce a novel message-exchanging optimization (MEO) model in [18] to optimize performance of distributed systems. In addition, we focus on the energy efficiency and we present the performance evaluation of our approach in this direction. Current efforts include various nodes communicating with each other in order to request for services by sending messages, however without considering the number of messages. We start with a related works section (Section 2) to present current approaches to and challenges in message exchanging in distributed systems. Based on this analysis, we conclude with a critical discussion of the key characteristics for messaging.

This work proposes the MEO model in order to optimize the energy efficiency of communication in distributed systems. The proposed solution forms a simple but significant approach in terms of both energy and performance efficiency. By not receiving negative requests from remote entities we optimize fundamental metrics, e.g. total delays. Nevertheless, this implies a variety of costs and concerns with regards to the size of messages, the definition of intervals, the topologies of the system and the throughput of jobs. To answer these issues we presented a MEO model that optimizes a number of time-centric performance criteria as well as energy-aware measures (e.g. the energy consumption rates based on the uptime of resources). The actual approach includes a mathematical representation of a graph theory model. To demonstrate the MEO approach in a real-case scenario we have implemented an inter-cloud simulation, where various services are submitted from users to meta-brokers for extracting resource availability. The simulation experiments draw a number of considerations as follows. (a) The diversity of message exchanging latencies shows increased performance in terms of energy consumption rates. (b) The collective model (operating in synchronous standards) optimizes the number of messages performance (e.g. for the configuration of the centralized experimental case the improvement factor is 3.5). (c) The ranking procedure is considered as first come first served fashion, and for this case the energy consumption levels are improved as well. (d) Both experimental cases show high adaptive-ness to various workloads and topologies. (e) The decentralization offers high dynamic-ness (e.g. for cases of low resource availability) by slightly affecting performance due to meta-brokering message exchanging delays.