Jacques Malenfant

From Data Center Resource Allocation to Control Theory and Back

Continuously adjusting the horizontal scaling of applications hosted by data centers appears as a good candidate to automatic control approaches allocating resources in closed-loop given their current workload. Despite several attempts, real applications of these techniques in cloud computing infrastructures face some difficulties. Some of them essentially turn back to the core concepts of automatic control: controllability, inertia of the controlled system, gain and stability. In this paper, considering our recent work to build a management framework dedicated to automatic resource allocation in virtualized applications, we attempt to identify from experiments the sources of instabilities in the controlled systems. As examples, we analyze two types of policies: threshold-based and reinforcement learning techniques to dynamically scale resources. The experiments show that both approaches are tricky and that trying to implement a controller without looking at the way the controlled system reacts to actions, both in time and in amplitude, is doomed to fail. We discuss both lessons learned from the experiments in terms of simple yet key points to build good resource management policies, and longer term issues on which we are currently working to manage contracts and reinforcement learning efficiently in cloud controllers.

Plan-based replication for fault-tolerant multi-agent systems

The growing importance of multi-agent applications and the need for a higher quality of service in these systems justify the increasing interest in fault-tolerant multi-agent systems. In this article, we propose an original method for providing dependability in multi-agent systems through replication. Our method is different from other works because our research focuses on building an automatic, adaptive and predictive replication policy where critical agents are replicated to avoid failures. This policy is determined by taking into account the criticality of the plans of the agents, which contain the collective and individual behaviors of the agents in the application. The set of replication strategies applied at a given moment to an agent is then fine-tuned gradually by the replication system so as to reflect the dynamicity of the multi-agent system

A Predictive Method for Providing Fault Tolerance in Multi-agent Systems

The growing importance of multi-agent applications and the need for a higher quality of service in these systems justify the increasing interest in fault-tolerant multi-agent systems. In this article, we propose an original method for providing dependability in multi- agent systems through replication. Our method is different from other works because our research focuses on building an automatic, adaptive and predictive replication policy where critical agents are replicated to avoid failures. This policy is determined by taking into account the criticality of the plans of the agents, which contain the collective and individual behaviors of the agents in the application. The set of replication strategies applied at a given moment to an agent is then fine-tuned gradually by the replication system so as to reflect the dynamicity of the multi-agent system. We report on experiments assessing the efficiency of our approach.

Experience and prospects for various control strategies for self-replicating multi-agent systems

Distributed cooperative applications (e.g.,e-commerce) are now increasingly being designed as a set of autonomous entities, named agents, which interact and coordinate(thus named a multi-agent system). Such applications are often very dynamic: new agents can join or leave, they can change roles, strategies, etc. This high dynamicity creates new challenges to the traditional approaches of fault-tolerance. As relative importance of agents may evolve during the course of computation and problem solving,we need to dynamically and automatically identify the most critical agents and to adapt their replication strategies (e.g., active or passive, number of replicas), in order to maximize their reliability and their availability. One important issue is then: what kind of information could be used to estimate which agents are most critical agents? In this paper, we will first introduce our prototype architecture for adaptive replication. Then, we will discuss various kinds of information and strategies to estimate criticality of agents: static dependences, dynamic dependences, roles, norms, and plans. Some preliminary measurements and future directions will also be presented.

Non-functional data collection for adaptive business processes and decision making

Monitoring application services becomes more and more a transverse key activity in SOA. Beyond traditional human system administration and load control, new activities such as autonomic management as well as SLA enforcement raise the stakes over monitoring requirements. In this paper, we address a new monitoring-based activity which is selecting among competitive service offers based on their currently measured QoS. Starting from this use case, the late binding of service calls in SOA given the current QoS of a set of candidate services, we first elicit the requirements and then describe M4ABP (Monitoring for Adaptive Business Process), a middleware component for monitoring services and delivering monitoring data to business processes wishing to call them. Within this settings, M4ABP provides solutions for general requirements: flexibility as well as performance in data access for clients, coherency of data sets and network usage optimization. Lessons learned from this first use case can be applied to similar monitoring scenario, as well as to the larger field of context-aware computing.

Composite contract enforcement in hierarchical component systems

Abadi and Lamport established a general theorem for composing specifications [1]. Based on an assume-guarantee principle, it enables one to prove the specification of a composite system from the ones of its components. But the general application of this theorem to software composition is not that straightforward because the resulting abstract specification of the composite depends upon the hidden guarantees of its subcomponents. In this paper, we investigate how this result can be exploited without blurring responsibilities between the different participants. Our contributions leverage an existing contracting framework for hierarchical software components [7], in which contracts are first-class objects during configuration and run times. This framework already associates specifications and responsibilities to software components, within the traditional horizontal form of composition. We show here how the vertical one can be made operational using the theorem as a sound formal basis. The resulting composite contracts make possible not only to detect violations, but also to determine and exploit precisely responsibilities upon them, related to both forms of composition.

TOWARDS A UNIFICATION OF SOME LINGUISTIC REPRESENTATION MODELS: A VECTORIAL APPROACH

This paper takes place in the computing with words framework where a unified model of several linguistic representation models is proposed. We consider (i) the 2-tuple fuzzy linguistic representation model, (ii) the proportional 2-tuple fuzzy linguistic representation model, (iii) the linguistic degrees and their associated generalized symbolic modifiers. Our approach is based on a vectorial representation. Using vectors and scalar multiplications, we translate and then unify these three models into a single one. In this paper, the scope of our research is limited to the definitions of a linguistic value, the attached aggregation operators remaining the next logical step.

Predictive fault tolerance in multiagent systems: a plan-based replication approach

The fact that multiagent applications are prone to the same faults that any distributed system is susceptible to and the need for a higher quality of service in these systems justify the increasing interest in fault-tolerant multiagent systems. In this article, we propose an original method for providing dependability in multiagent systems through replication. Our method is different from other works because our research focuses on building an automatic, adaptive and predictive replication policy where critical agents are replicated to minimize the impact of failures. This policy is determined by taking into account the criticality of the plans of the agents, which contain the collective and individual behaviors of the agents in the application. The set of replication strategies applied at a given moment to an agent is then fine-tuned gradually by the replication system so as to reflect the dynamicity of the multiagent system. Some preliminary measurements were made to assess the efficiency of our approach and future directions are presented.

Composition and compositionality in a component model for autonomous robots

Component models for autonomous robots control architectures are much more constrained than traditional ones: obeying strict timing constraints, coping with a large spectrum of rapidly changing hardware (e.g. sensors and actuators), etc. Beyond introducing new concepts into components themselves, composition in such models must go much farther than the standard connection through method signature interfaces. Viewing components as full-fledged sensori-motor behaviors, our model follows the concept of rich interfaces introduced by Henzinger et al. to attach to each component all the necessary syntactical and behavioral information to make them externally composable. This paper presents two kinds of composition, parallel and by modes, their semantics, their compositionality properties and the impact of these on the composition model. A prototype implementation in Java is backed by a constructive semantics defined as a constraint system solved in this prototype with the ECLIPSe constraint programming system.

A Token-Based Scheme for Coordinating Decisions in Large-Scale Autonomic Systems

Coordinating decisions in large-scale systems is acknowledged as one of the most challenging aspects of autonomic computing. If control theory is already pushed to its limits by the inevitable decentralised nature of such systems, going large-scale makes things even harder. In this paper, we propose a decentralised token-based scheme for coordinating shared resources allocation in large-scale autonomic systems. In this scheme, each token represents an exclusive access to a share of the resource. Tokens are used and exchanged among autonomic managers to limit their decisions to the resource capacity the tokens they hold at this moment represent. Our token based protocol is meant to provide statistical guarantees on the average total resource usage and the average lateness of node actions due to the coordination. It exposes a set of parameters which we have explored to characterise its fairness and global resource usage on a use case in geotracking. Experiments with the coordination of 10.000 autonomic managers have shown very good results for large spectrum of parameter values and systems regimes.