Rucon - Runtime Control on Multi Clouds

Cloud computing is a disruptive technology for global on demand computing, which facilitates the delivery of computational resources as services. Cloud operating systems (Cloud OSs) manage the execution of underlying resources, e.g., Physical Machines (PM) or Virtual Machines (VM), in compliance with user constraints. An application is typically deployed on a single Cloud. Nowadays, we are faced with the complex ultra scale socio-lifestyle applications like applications for the management of smart cities or for the medical body area networks that are being deployed on Clouds. Such applications have demanding user requirements and need diversity of available IT infrastructures to be deployed. Moreover, economical and ecological reasons lead to the utilization of multiple Clouds for the deployment of a single application. This concept where a Service Provider (SP) is responsible for handling the additional complexity of coordinating multiple services across multiple external Infrastructure Providers (IPs) is called multi Cloud. While there are mature methods for guaranteeing (Quality of Service) QoS in a single Cloud, guaranteeing QoS of an application deployed in multi Clouds is currently an unsolved scientific question. In the recent years autonomic computing emerged as a promising solution for the management of complex computing systems by minimizing human intervention into the system. The objective of this project is to develop autonomic computing methodologies and tools for application agnostic guaranteeing of QoS in multi Clouds. Based on the complexity of the multi Cloud, the type and quality of data used for the monitoring of the QoS, and the ability or inability to create the behavioral model or an application we will investigate the most prominent autonomic mechanisms that range from heuristics, control-theoretic to the machine learning based approaches. Thus, instead of making the whole application self-adaptive, we make each tier of the application, that can basically be deployed at different IPs self-adaptive. Then, we devise a coordination methodology for maintaining QoS of multiple tiers deployed at different IPs.

Nowadays large data centers are used to host applications that range from social media to video streaming apps. Meanwhile our society is dependent on the availability of those services for online teaching or to maintain smart buildings. However, such data centers require a significant amount of resources to be operated properly including energy, water for cooling, space, and human resources. In the past the research focus was merely on the optimization of the performance of the data centers and less on the optimization of the resource usage. With upcoming, even more energy hungry generative AI, competing issues of reliability, fault tolerance and sustainability have been raised. Higher reliability of ICT resources (i.e., data centers) usually cause higher utilization of resources. Only recently researchers discovered that generating a single picture with generative AI consumes the same amount of energy as charging a smartphone. Thus, a conflicting research issue appeared during the last years addressing the necessity to deliver reliable and fault tolerant systems while reducing resource consumption and consequently maintaining sustainability. In this project we propose the paradigm shift from the traditional (deterministic) methods for resource utilization in large scale data centers and propose novel data driven (non-deterministic) methods that deliver reliability and fault-tolerance while maintaining certain sustainability goals. We utilize the concept of Multi Clouds where geographically distributed data centers work jointly to achieve certain performance metrics by being able to balance the varying workload and possible failures of the system. With the concept of Edge computing we preprocess the data along the data transmission pipeline through infrastructure available (like routers, switches and micro data centers), repair data of low quality and ensure high quality of the decision making at the massive Clouds. We focus in particular on the stateless problem of data analytics when decisions are made on old AI models that do not represent the reality any more in the accurate manner. We measure and control staleness for distributed online analytics applications with extensive theoretical and empirical evaluations and propose novel methods to ensure freshness of the model without compromising sustainability. Based on our deep understanding of system's behavior we gained in the Rucon project, we opened two new research directions, namely on hybrid classic-quantum systems where resources in data centers are partially allocated on quantum machines. This allows us to achieve sustainability with completely different orders of magnitude compared to traditional systems. Another research direction is Computational Sustainability that considers the development of methods to (i) increase sustainability in IT infrastructures and to (ii) utilize IT tools and methods to make a better world and - for example - to combat climate change.