Workshop Program

Integrating solar and wind generation into the electrical grid, developing the next generation of mobile communications infrastructure, and adaptively controlling the traffic flow in a large city at rush hour all have a common challenge: integration of large and variable numbers of independent heterogeneous devices, some of which may not even exist at design time.  For such complex and interconnected systems, complexity itself is a central technical challenge.

The natural world provides a wealth of examples of ensembles of heterogeneous components which collectively exhibit determinate and reproducable behavior.  These examples demonstrate that self-organization is possible, but it is not yet clear how best to achieve such results in our engineered systems.  Results from our testbed of small robotic vehicles suggest that the properties and capabilities appropriate for integrating a few agents may be significantly different from those that enable desired self-organized behaviors of ensembles of large numbers of agents.

Self-adaptation in embedded self-organizing real-time systems pose stringent expectations on the performance of their system architecture. The flexibility required for self-adaptation argues for multiplicity or variability of processes, whereas the embedded real-time aspects argue for very fast and low power processes. These two expectations are in direct opposition to each other, and good engineering practice implies careful study of the implications of the trade-offs. In this paper, we show how to use our Wrapping approach to integration infrastructure as a base to study proposed infrastructure choices for these applications, and argue that the Wrappings approach is ideally suited to this endeavor, since it makes no a priori assumptions about the infrastructure (or about itself, as we shall explain), and therefore allows any such questions to be studied. To do so, we provide enough details about the Wrapping approach to support our claims, and show how it would be applied to some popular infrastructures.

The Artificial Hormone System (AHS) is a completely decentralized operation principle for a middleware which can be used to allocate tasks in a system of heterogeneous processing elements (PEs) or cores. Tasks are scheduled according to their suitability for the heterogeneous PEs, the current PE load and task relationships. The AHS also provides properties like self-configuration, self-optimization and self-healing in the context of task allocation. In addition, it is able to guarantee real-time bounds for such self-X-properties. 

The operation principle of the AHS is based on the hormone loop. This is a sequence of actions and wait states executed periodically on each PE. We present a formal specification of the hormone loop in this paper. The outcome is to guarantee consistent hormone computation (important for holding the real-time bounds of the self-X properties) and a fast recognition of task or PE failures. Even more, we present an algorithm to terminate single PEs consistently.

Decision making in a self-optimising distributed Organic Computing system requires information about the system’s state. Accurate information enables the overall system to respond better to state changes. Distributed systems can use different network protocols, e.g UPD or TCP/IP, to connect the nodes. There is no guarantee that all of these network protocols are able to send broadcasts. If all used network protocols support broadcast it is still not sure that broadcasts can be sent across different protocol domains, e.g. from UPD to TCP/IP. We use therefore distributed hash tables (DHT) to enable an application layer broadcast for information dissemination, which only sends unicast messages in the network layer to spread node status information in a distributed system. In DHTs the node IDs are used to determine the communication partner. The node IDs are generated randomly in DHTs. In this paper we show how choosing IDs systematically, instead of generating them randomly, influences the network usage by using our DHT broadcast algorithms for information dissemination.

Several applications of Organic Computing (OC) systems as well as Autonomic Computing (AC) systems are based on self-optimising multi-agent systems, i.e. distributed autonomous devices. One of the main challenges of these is to emerge towards a global optimal system state based only on local information for each agent. In order to reach a global optimal state some agents need to avoid selfish actions and instead consider the benefits of their actions for the whole system. Choosing the action of an agent is often based on solving optimisation problems, which can be modelled as a distributed constraint optimisation problem (DCOP). This paper presents a new asynchronous approach to solve DCOP by extending only the underlying local objective function of each agent. The main benefit of this approach is the avoidance of an additional complex decision making algorithm that may interfere with the original task of an agent and reduces the scalability of the system. Exemplary, a distributed constraint optimisation problem is considered to quantify the effectiveness and computation as well as communication cost of the discussed approach.

With the growing use of smartphones and other mobile devices, it becomes essential to be able to assure the user that his system and applications are doing exactly what they are supposed to do. Over the years and despite its configuration complexity, Mandatory Access Control has proven its efficiency in protecting systems. This paper proposes a solution providing a generic protection that doesn’t need to modify the applications. Moreover, in order to face the complexity of defining an efficient MAC policy, a tool automatizes the generation of the policies required for the various applications.

However, to efficiently guarantee the security of a system, each layer that composes it must be secured. Therefore, MAC implementations should not be limited to the operating system, but should also protect the inside of the applications.

This paper presents Security Enhanced Dalvik (SEDalvik), a MAC approach for the Dalvik Virtual Machine in order to control the flows inside the Java applications running in Android.

SEDalvik proposes a new mandatory protection to block the attacks that exploit the weakness of the Dalvik VM. By controlling the information flows between the Java objects, SEDalvik could prevent the new vectors of attack coming from the threat of the Java virtual machine as explained by Kaspersky Labs1. In contrast with other approaches, our solution corresponds to a self-organizing system since it transparently protects existing Java applications without any modifications. An experiment on an Android phone shows the efficiency of the protection.