Search

Cloud applications are often subject to unexpected events like flashcrowds and hardware failures. Users that expect a predictable behavior may abandon an unresponsive application when these events occur. Researchers and engineers addressed this problem on two separate fronts: first, they introduced replicas - copies of the application with the same functionality - for redundancy and scalability; second, they added a self-adaptive feature called brownout inside cloud applications to bound response times by modulating user experience. The presence of multiple replicas requires a dedicated component to direct incoming traffic: a load-balancer. Existing load-balancing strategies based on response times interfere with the response time controller developed for brownout-compliant applications. In fact, the brownout approach bounds response times using a control action. Hence, the response time, that was used to aid load-balancing decision, is not a good indicator of how well a replica is performing. To fix this issue, this paper reviews some proposal for brownout-aware load-balancing and provides a comprehensive experimental evaluation that compares them. To provide formal guarantees on the load balancing performance, we use a randomized optimization approach and apply the scenario theory. We perform an extensive set of experiments on a real machine, extending the popular lighttpd web server and load-balancer, and obtaining a production-ready implementation. Experimental results show an improvement of the user experience over Shortest Queue First (SQF)-believed to be near-optimal in the non-adaptive case. The improved user experience is obtained preserving the response time predictability.

In this paper, we address distributed convergence to fair allocations of CPU resources for time-sensitive applications. We propose a novel resource management framework where a centralized objective for fair allocations is decomposed into a pair of performance-driven recursive processes for updating: (a) the allocation of computing bandwidth to the applications (resource adaptation), executed by the resource manager, and (b) the service level of each application (service-level adaptation), executed by each application independently. We provide conditions under which the distributed recursive scheme exhibits convergence to solutions of the centralized objective (i.e., fair allocations). Contrary to prior work on centralized optimization schemes, the proposed framework exhibits adaptivity and robustness to changes both in the number and nature of applications, while it assumes minimum information available to both applications and the resource manager. We finally validate our framework with simulations using the TrueTime toolbox in MATLAB/Simulink.

A previous paper addressed the "PI+p" controller (a PI augmented with an additional, stable pole) and demonstrated its advantages over standard PIs, particularly if tuned with a combined model-relay-based method along the "contextual" approach proposed in another recent work. This manuscript extends the scenario to the PID+p structure, with obvious meaning, and applies the contextual approach to its autotuning. Benchmark tests and a quite realistic simulation example support the proposal.

The equation-based Modelica language allows the modeller to specify custom functions. The body of a function is an algorithm that contains procedural code to be executed when the function is called. This language feature is useful for many applications; however, the insertion of a function often prevent model optimizations that require the model to be formulated in purely declarative form by equations only. This paper discusses several non-trivial cases in which the function call and the corresponding algorithmic code can be transformed into an equivalent purely equation-based model, thus allowing further optimization. The inlining algorithms presented in the paper go well beyond the state of the art in commercial and open-source Modelica tools. (Less)

This paper presents a library of (PID) controller models adopting the object-oriented approach, and written in Modelica. Peculiar to this work is that controllers are represented both as dynamic continuous-time and digital models, achieving consistence between the two and accounting for the functionalities of typical industrial implementations. This allows the designer to use realistic controllers, maintaining the possibility of choosing the continuous-time or digital (event based) representation. The former allows for example for variable-step simulation, to the advantage of efficiency, while the latter represents very realistically the actual control system's operation, clearly at the cost of more simulation time. Beside standard PI and PID controls, in this work also autotuning is (initially) considered, and some application examples are reported to show how the presented library can ease system studies involving (PID) controls.

The application of feedback control to computing systems is a promising research area, but has to date been hindered by the almost unanimously perceived complexity in creating control-oriented system models. Computing systems are in fact considered very hard to describe with dynamic models allowing for simple and powerful control design tools, so that complex ones need bringing in to the detriment of efficiency and result assessment. In this work a novel approach to the modelling of computing systems is proposed, in a view to explain and partially avoid such complexity, by capturing only their relevant dynamics with the simplest possible models. The approach is shown to work at least on two relevant case studies, so that a significant generality can be inferred from it, being able to reproduce the relevant parts of the system's behaviour and paving the way to control design and synthesis. (Less)

In some recent papers it was shown that preemptive process schedulers in multitasking operating systems can be viewed, and above all designed , as discrete-time feedback controls with very simple (I- or PI-type) regulators, yielding significant advantages over classical scheduling policies as for time complexity and parameter interpretability. In this work, the same problem is tackled with a predictive control approach. Doing so allows to release some simplification hypotheses of the mentioned papers and to improve the achieved solutions’ flexibility, widening the application possibilities at an affordable additional cost.

A recently published paper addressed the so called “PI+p” controller (i.e., a PI augmented with an additional, stable pole) and demonstrated that said controller yields interesting results particularly if tuned with the “contextual” approach, where model-based rules are used in conjunction with a relay experiment so as to provide at the same time the tuned regulator, and the tuning model. In this manuscript the scope is extended to the “PID+p” (with obvious meaning) controller structure, discussing its potentialities with reference to some benchmark cases.

This manuscript is part of a long-term research, aimed at establishing methodologically grounded relationships between model- and relay-based tuning of industrial regulators, and at consequently deriving synthesis procedures that couple the advantages of model-based tuning to the simplicity and clarity of relay-based identification. In this work, the addressed controller structure is the “PI+p”, i.e., a PI augmented with an additional, stable pole. The advantages of using the combined model-relay-based approach on that structure are evidenced, by means of both simulation and experimental results.