[Context and Motivation] Global energy consumption has been steadily increasing in recent years, with data centers emerging as major contributors. This growth is largely driven by the widespread migration of applications to the Cloud, alongside a rising number of users consuming digital content. Dynamic adaptation (or self-adaptive) approaches appear as a way to reduce, at runtime and under certain constraints, the energy consumption of software applications. [Question/Problem] Despite efforts to make energy-efficiency a primary goal in the dynamic adaptation of software applications, there is still a gap in understanding how to equip these self-adaptive software systems (SAS), which are dynamically adapted at runtime, with effective energy consumption monitoring tools that enable energy-awareness. Furthermore, the extent to which such an energy consumption monitoring tool impacts the overall energy consumption of the SAS ecosystem has not yet been thoroughly explored. [Methodology] To address this gap, we introduce the EnCoMSAS (Energy Consumption Monitoring for Self-Adaptive Systems) tool that allows to gather the energy consumed by distributed software applications deployed, for instance, in the Cloud. EnCoMSAS enables the evaluation of energy consumption of SAS variants at runtime. It allows to integrate energy-efficiency as a main goal in the analysis and execution of new adaptation plans for the SAS. In order to evaluate the effectiveness of EnCoMSAS and investigate its impact on the overall energy consumption of the SAS ecosystem, we conduct an empirical study by using the Adaptable TeaStore case study. Adaptable TeaStore is a self-adaptive extension of the TeaStore application, a microservice benchmarking application. For this study, we focus on the recommender service of Adaptable TeaStore. Regarding the experiments, we first equip Adaptable TeaStore with EnCoMSAS. Next, we execute Adaptable TeaStore by varying workload conditions that simulate users interactions. Finally, we use EnCoMSAS for gathering and assessing the energy consumption of the recommender algorithms of Adaptable TeaStore. To run these experiments, we use nodes of the Grid5000 testbed. [Results] The results show that EnCoMSAS is effective in collecting energy consumption of software applications for enabling dynamic adaptation at runtime. The observed correlation between CPU usage and energy consumption collected by EnCoMSAS provides evidence supporting the validity of the collected energy measurements. Moreover, we point out, through EnCoMSAS, that energy consumption is influenced not only by the algorithmic complexity but also by the characteristics of the deployment environment. Finally, the results show that the impact of EnCoMSAS on the overall energy consumption of the SAS ecosystem is comparatively modest with respect to the entire set of the TeaStore applications microservices.


翻译:[背景与动机] 近年来全球能源消耗持续增长,数据中心已成为主要能耗源之一。这一增长主要源于应用程序向云端的大规模迁移,以及消费数字内容的用户数量不断增加。动态自适应方法似乎提供了一种途径,可在运行时及特定约束条件下降低软件应用的能耗。[问题陈述] 尽管已有研究尝试将能效作为软件应用动态适配的主要目标,但对于如何为运行时动态调整的自适应软件系统配备有效的能耗监测工具以实现能耗感知,仍存在认知空白。此外,此类能耗监测工具对SAS生态系统整体能耗的影响程度尚未得到充分探究。[方法论] 为填补这一空白,我们提出了EnCoMSAS(面向自适应系统的能耗监测工具),该工具能够收集部署于云端等环境的分布式软件应用的能耗数据。EnCoMSAS支持在运行时评估SAS变体的能耗表现,并可将能效作为SAS新适配方案分析与执行的核心目标。为评估EnCoMSAS的有效性并探究其对SAS生态系统整体能耗的影响,我们采用自适应TeaStore案例开展实证研究。自适应TeaStore是TeaStore应用(一种微服务基准测试应用)的自适应扩展版本。本研究聚焦于自适应TeaStore的推荐服务。实验过程中,我们首先为自适应TeaStore集成EnCoMSAS工具,随后通过模拟用户交互的差异化工作负载条件执行系统运行,最终利用EnCoMSAS收集并评估自适应TeaStore推荐算法的能耗数据。所有实验均在Grid5000测试平台的节点上运行。[结果分析] 结果表明,EnCoMSAS能有效收集软件应用的能耗数据以支持运行时动态适配。通过EnCoMSAS观测到的CPU使用率与能耗数据之间的相关性,为所采集能耗测量值的有效性提供了证据支持。此外,我们借助EnCoMSAS指出:能耗不仅受算法复杂度影响,还取决于部署环境的特性。最终数据显示,相较于TeaStore应用微服务整体集合,EnCoMSAS对SAS生态系统总能耗的影响相对有限。

0
下载
关闭预览

相关内容

DeepSeek模型综述:V1 V2 V3 R1-Zero
专知会员服务
116+阅读 · 2月11日
Pytorch多模态框架MMF
专知
50+阅读 · 2020年6月20日
国家自然科学基金
9+阅读 · 2015年12月31日
国家自然科学基金
46+阅读 · 2015年12月31日
国家自然科学基金
3+阅读 · 2015年12月31日
国家自然科学基金
0+阅读 · 2014年12月31日
VIP会员
相关基金
国家自然科学基金
9+阅读 · 2015年12月31日
国家自然科学基金
46+阅读 · 2015年12月31日
国家自然科学基金
3+阅读 · 2015年12月31日
国家自然科学基金
0+阅读 · 2014年12月31日
Top
微信扫码咨询专知VIP会员