SPANDO · Self-organizing Performance Prediction and Optimization for Large-scale Software Systems

Self-organizing Performance Prediction and Optimization for Large-scale Software Systems

The SPANDO project - acronym for Self-organising Performance Prediction and Optimisation for Large-scale Software Systems - focuses on the use of run-time performance prediction models and self-organising adaptation strategies to achieve performance optimisation of large-scale computing systems. Large-scale complex systems are becoming popular in many different areas of computer systems, spanning from cloud computing systems to decentralised mobile platforms. Most of the research efforts of SPANDO focused on cloud computing, since it has become popular in organisations as a mean to outsource part of their IT infrastructures and lower their costs. As the number of cloud providers and offerings are increasing, it is becoming more challenging for organizations to make decisions on which cloud resources to rent for deploying their applications. Cloud resources are not only characterized by different performance in terms of processors, memory, etc., but also by different pricing models that result in different levels of reliability. For example, Amazon EC2 spot resources and Google CE’s preemptible resources are rented at a cheaper price than on-demand resources, but they carry the continuous risk of being claimed back by the cloud provider. While these types of resources operate at lower costs than others, they introduce new issues since they can become unavailable at any time. Within SPANDO we create models to understand how these systems evolve, and use these models to find system reconfigurations that allow to minimize the operating costs while guaranteeing the desired level of quality. Beside cloud computing systems, SPANDO also provided investigations on additional contexts, such us performance optimization in decentralised voluntary/edge computing systems. Description of the work performed The work of SPANDO has been articulated within four key activities: performance prediction, performance optimisation, development of software abstractions, and real case studies evaluation. In the performance prediction activity, we have developed a prediction model, usable both at design-time and run-time, which calculates the average response time and response time distribution of a cloud computing system composed of different application components and different resources. The main innovation compared to the state of the art is that we consider the possibility to use cheaper problematic resources (e.g., unreliable preemptive resources and burstable resources with variable speed), and that our approach is designed to be self-adaptive. The performance optimisation activity resulted in a self-organization heuristic called OptiSpot, for optimising the cost and the performance of a cloud application using the performance prediction model mentioned above. In this activity we have also explored a decentralised self-organisation performance optimisation technique applied to voluntary/edge clouds called MycoCloud. Software abstractions have been developed as a collection of tools called SpandoTools. These tools offer a way to easily use the prediction and adaptation models developed in SPANDO in possible commercial applications or in other research projects. Finally, data from real case studies and standardised benchmarks such as SAP ERP and SPECjAppServer have been used to validate the models and the tools released. Description of the main results The main results of SPANDO have been delivered in the form of academic publications containing a detailed description of the conceptual framework, models, and heuristics used within the project. The prediction and optimisation algorithms have been released as a collection of software tools. Possible ways for using these tools have been documented in the accompanying conference and journal publications, as well within the code base. Thanks to the tools we have developed in SPANDO, application developers have now the possibility to adjust the deployment of their cloud applications by taking advantage of the full range of resources currently offered by the most popular cloud providers instead of just using the most expensive on-demand ones. According to experiments using data from real-world case studies, a cloud application optimised with SpandoTools may have its deployment costs reduced up to 90% when compared to the use of deterministic homogeneous on-demand resources. Impact SPANDO scientific impact has been demonstrated by the dissemination of the results through seven publications in internationally peer-reviewed venues. One of these publications has also received a prestigious best paper award. SPANDO has promoted international networking and exported its techniques to several other institutions through direct project-funded collaborations with the College of William and Mary (Williamsburg, VA, USA), Gran Sasso Science Institute (L’Aquila, Italy), Politecnico di Milano (Milan, Italy), Fondazione Bruno Kessler (Trento, Italy), and MIT Media Lab (Cambridge, MA, USA). The project also included dissemination and outreach activities that go beyond collaborations. Presentations for the project results have been given at BBC Research (London, UK), Gran Sasso Science Institute (L’Aquila, Italy), Northeastern University (Boston, MA, USA), and several internal outreach events organised within Imperial College London. Dissemination and collaborations are continuing after the end of the project through the project website, which has made all the project publications and software modules available for download with open access. Since the project has left several possibilities for further developments, we expect that all the available material will: (i) ease the adoption of SPANDO techniques by cloud software companies to reduce their costs and improve the experience of their customers; and (ii) ease the creation of follow-up projects in the research community. For more information regarding SPANDO, its documentation and its tools, visit the official website at http://www.spando.org or contact Dr. Daniel J. Dubois (daniel@dubois.it).

Текст от CORDIS, на английски · Данни: CORDIS, © Европейски съюз