Abstract Autoscaling enables container cluster orchestrators to automatically adjust computational resources, such as containers and Virtual Machines (VMs), to handle fluctuating workloads effectively. This adaptation can involve modifying the amount of resources (horizontal scaling) or adjusting their computational capacity (vertical scaling). The motivation for our work stems from the limitations of previous autoscaling approaches, which are either partial (scaling containers or VMs, but not both) or excessively complex to be used in real systems. This complexity arises from their use of models with a large number of variables and the addressing of two simultaneous challenges: achieving the optimal deployment for a single scheduling window and managing the transition between successive scheduling windows. We propose an Integer Linear Programming (ILP) model to address the challenge of autoscaling containers and VMs jointly, both horizontally and vertically, to minimize deployment costs. This model is designed to be used with predictive autoscalers and be solved in a reasonable time, even for large clusters. To this end, improvements and reasonable simplifications with respect to previous models have been carried out to drastically reduce the size of the resource allocation problem. Furthermore, the proposed model provides an enhanced representation of system performance in comparison to previous approaches. A tool called Conlloovia has been developed to implement this model. To evaluate its performance, we have conducted a comprehensive assessment, comparing it with two heuristic allocators with different problem sizes. Our findings indicate that Conlloovia consistently demonstrates lower deployment costs in a significant number of cases. Conlloovia has also been evaluated with a real application, using synthetic and real workload traces, as well as different scheduling windows, with deployment costs approximately 20% lower than heuristic allocators.

For a long time, the common billing time slot used by cloud providers was the hour, but recently they have changed it to one second with a minimum charge of one minute. In this paper, the impact of this change on virtual machine allocation strategies is analyzed. With the minimum one-minute charge, the state-of-the-art allocation strategies are no longer optimal. This paper proposes a new analytic model to obtain the optimal allocation that minimizes the total cost considering this new billing scheme. However, the optimization problem using this model is unmanageable due to its computational complexity. Therefore, other sub-optimal allocations strategies are proposed. The performance of these strategies (in terms of total allocation cost and computational effort) is analyzed in order to assess the influence of the time slot length. Allocation time slots of one hour, one minute and one second are compared. The experimental work is carried out using synthetic workloads based on two real public traces. The study concludes that the time slot of one minute offers the best trade-off between allocation cost and computational effort. The experimentation shows that the proposed strategy can save up to 14 percent over using the time slot of one hour.

Summary The advent of hybrid cloud technologies and public Infrastructures as a Service (IaaS) allows service developers to offer services to their customers with little upfront investment and to adapt services to different workload sizes. The problem of minimizing the costs of the hired public infrastructure while providing the quality of service needed by the final customer arises when using hybrid clouds. Several scheduling strategies have been proposed to solve this problem for services dealing with deadline‐constrained bag‐of‐tasks workloads. Most of these solutions do not consider the variable performance of the clouds, the provisioning delay of virtual machine instances that affects the elasticity, and the impracticality of having good processing time estimations in real systems. We propose a scheduler algorithm that overcomes previous limitations and can minimize the cost of the infrastructure while maximizing the number of deadlines met by the service. Our solution can work autonomously by using sampled observations of the processing times and considers the heterogeneity and the provisioning time of the virtual machine instances. An evaluation was conducted by simulating different scenarios and workload types. Simulation results show that our solution obtains better or similar results than previous techniques in most scenarios in terms of deadlines met.

Cluster management has become a multi objective task that involves many disciplines like power optimization, fault tolerance, dependability and online system operation analysis. Efficient and secure operation of these clusters is a key objective of any data center policy. In addition, the service provided by these servers must fulfill a level of quality of service (QoS) to the customers. Applying self-management techniques to these clusters would simplify and automate its operation. Current self-management techniques that take into account service level agreements (SLAs) do not cover at the same time the two most important sides of the cluster operation: self-optimization, for an efficient and profitable operation, and self-healing, for a secure operation and high level of quality of service perceived by users. This work integrates a self-optimization strategy for Internet server clusters that optimizes the power consumption, using dynamic provisioning of servers, with a self-healing strategy that improves the reaction of the cluster to a server failure, by using the spare capacity of the cluster intelligently. The self-management technique is based on empirical response time and power consumption models of the servers that simplify its operation. Additionally, the technique presented in this paper guarantees the fulfillment of the SLA.

Nowadays, enterprises providing services through Internet often require online services supplied by other enterprises. This entails the cooperation of enterprise servers using Web services technology. The service exchange between enterprises must be carried out with a determined level of quality, which is usually established in a service level agreement (SLA). However, the fulfilment of SLAs is not an easy task and requires equipping the servers with special control mechanisms which control the quality of the services supplied. The first contribution of this research work is the analysis and definition of the main requirements that these control mechanisms must fulfil. The second contribution is the design of a control mechanism which fulfils these requirements and overcomes numerous deficiencies posed by previous mechanisms. The designed mechanism provides differentiation between distinct categories of service consumers as well as protection against server overloads. Furthermore, it scales in a cluster and does not require any modification to the system software of the host server, or to its application logic.

Stochastic analysis techniques for real-time systems model the execution time of tasks as random variables. These techniques constitute a very powerful tool to study the behaviour of real-time systems. However, as they can not avoid all the timing bugs in the implementation, they must be combined with measurement techniques in order to gain more confidence in the implemented system. In this paper, a set of tools to measure, analyze and visualize traces of real-time systems is presented. These tools are driven by stochastic models. In order to find bugs in the timing behaviour of the system, two metrics, called "pessimism" and "optimism", are proposed. They are based on two random variables, the optimistic and the pessimistic execution time, which are also introduced in this paper. These metrics are used in the debugging tools to compare the model and the measured system in order to find errors. The metrics are examined in three case studies.