Impact of Dynamics on Situated and Global Aggregation Schemes

Abstract

Abstract. Recently, numerous management approaches have emerged in order to manage networks and services in a decentralized and autonomous way. Some of them propose to minimize their cost by using a situated view when collecting aggregates for the decision making process, while others propose to improve their accuracy by using a more conventional approach which is global view. So far, little attention is given to the evaluation of situated view while many studies propose to evaluate global approaches. As a consequence, there is no work in the literature that compares the performance of situated and global aggregation schemes. Being able to choose the suitable approach for a given context is still a real challenge. Mastering it will ensure the efficiency of the autonomous management system. In this paper, we present a comparative study of situated and global schemes deployed over large scale and dynamic networks. We consider two factors: network and information dynamics. We implement typical aggregation schemes from each category and then we compare them according to the accuracy of the estimated aggregates and the efficiency of the decision making process.

SAAM: A self-adaptive aggregation mechanism for autonomous management systems

“…In summary, these results confirm our expectations and they are consistent with those obtained through simulations in [4] showing the validity of our fuzzy model.…”

“…The latter infers the current network operational state and dynamically decides which aggregation approach should be used and/or what changes must occur in it. SAAM considers typical situated and global aggregation schemes that we evaluated in [4]. These are: (1) a push-pull gossip scheme inspired from [5], where at each round, each node contacts one neighbor chosen uniformly at random and exchanges information with it ; (2) a push tree-based scheme based on GAP [6], where nodes communicate their local aggregates to a single root node that computes an overall aggregate and spreads it on all the interested nodes through a publishsubscribe mechanism ; and finally (3) a pull-based situated scheme based on the membership protocol HyParView [7] and Propagate2All [8], where each node maintains a view of its h-hops neighbors.…”

“…The NSM collects information about the Network Dynamics (ND), Network Size (NS) and aggregated data dynamics (ID: Information Dynamics). These metrics have a direct impact on the cost and performance of aggregation [4].…”

“…• Fuzzification: the CPE inputs/outputs are converted into fuzzy variables. The fuzzy sets of each variable are de-fined according to the performance evaluation presented in [4] ( Figure 2). The set of membership functions we defined for FLC variables are Very Low (VL), Low (L), Medium (M), High (H) and very high (VH).…”

“…To evaluate the performance of SAAM, we rely on the FreePastry simulator, which enables the creation of the network nodes and the messages exchanges. Concerning the tree-based scheme, we used Scribe [4] to maintain a multicast tree for the diffusion of the global aggregate to all nodes that monitor it. Simulation parameters are illustrated in Table III.…”