Catálogo de publicaciones - libros

With an increasing number of devices that must be managed, the scalability of network and service management is a real challenge. A similar challenge seems to be solved by botnets which are the major security threats in today’s Internet where a botmaster can control several thousands of computers around the world. This is done although many hindernesses like firewalls, intrusion detection systems and other deployed security appliances to protect current networks. From a technical point of view, such an efficiency can be a benefit for network and service management. This paper describes a new management middleware based on botnets, evaluates its performances and shows its potential impact based on a parametric analytical model.

Hierarchical event correlation is very important for distributed monitoring network and distributed system operations. In many large-scale distritbuted monitoring environments such as monitions senor networks for data aggregation, battlefield compact operations, and security events, an efficient hierarchical monitoring agent architecture must be constructed to facilitate event reporting and correlation utilizing the spacial relation between events and agents with minimum delay and cost in the network. However, due to the significant agent communication and management overhead in organzine agents in distributed monitoring, many of the existing approaching become inefficient or hard to deploy. In this paper, we propose a topology-aware hierarchical agent architecture construction technique that minimizes the monitoring cost while considering the underlying network topology and agent capabilities. The agent architecture construction is performed in a purely decentralized fashion based on the agents’ local knowledge with minimal communication and no central node support.

Service providers are constantly seeking ways to reduce the costs incurred in managing the services they deliver. With the increased distribution and virtualization of resources in the next generation network infrastructure, novel resource management approaches are sought for effective service delivery. In this paper, we propose a market-based hierarchical resource management mechanism using Machine Learning, which consists of a negotiation phase where customers are allocated the resources needed by their activated service instances, and a learning phase where service providers adjust the prices of their resources in order to steer the network infrastructure towards the desired goal of increasing their revenues, while delivering the mix of services requested by their customers. We present the operation of such a market where distributed and virtualized resources are traded as commodities between autonomic resource brokers performing the negotiation and learning on behalf of service providers. We perform extensive simulations to study the performance of the proposed hierarchical resource management mechanism.

Past research on probing-based network monitoring provides solutions based on preplanned probing which is computationally expensive, is less accurate, and involves a large management traffic. Unlike preplanned probing, adaptive probing proposes to select probes in an interactive manner sending more probes to diagnose the observed problem areas and less probes in the healthy areas, thereby significantly reducing the number of probes required. Another limitation of most of the work proposed in the past is that it assumes a deterministic dependency information between the probes and the network components. Such an assumption can not be made when complete and accurate network information might not be available. Hence, there is a need to develop network monitoring algorithms that can localize failures in the network even in the presence of uncertainty in the inferred dependencies between probes and network components. In this paper, we propose a fault diagnosis tool with following novel features: (1) We present an adaptive probing based solution for fault diagnosis which is cost-effective, failure resistant, more accurate, and involves less management traffic as compared to the preplanned probing approach. (2) We address the issues that arise with the presence of a non-deterministic environment and present probing algorithms that consider the involved uncertainties in the collected network information.

Our research aims to assist operators in finding solutions for faults using distributed case-based reasoning. One key operation of the distributed case-based reasoning system is to retrieve similar faults and solutions from various online knowledge sources. In this paper, we propose a multi-vector representation method which employs various and vectors to exploit the characteristics of faults described in semi-structured data. Experiments show that this method performs well in fault retrieval.

Fault detection is a crucial issue in autonomic networks for identifying unreliable nodes and reducing their impact on the network availability and performance. We propose in this paper to improve this situation based on the concept of promised cooperation. We exploit the promise theory framework to model voluntary cooperation among network nodes and make them capable of expressing the trust in their measurements during the detection process. We integrate this scheme into several distributed detection methods in the context of ad-hoc networks implementing the OLSR routing protocol. We quantify how the fault detection performances can be increased using this approach based on an extensive set of experimentations performed under the ns-2 network simulator.

Monitoring is an inherent part of the management loop. This paper studies the problem of quantifying utility of monitoring in a fully distributed load balancing setting. We consider a system where job requests arrive to a collection of identical servers. The goal is to provide the service with the lowest possible average waiting time in a fully distributed manner (to increase scalability and robustness).

We present a novel adaptive load balancing heuristic that maximizes utility of information sharing between the servers. The main idea is to forward the job request to a randomly chosen server and to collect load information on the request packet as it moves on. Each server decides, based on that information, whether to forward the job request packet to another server, or to execute it locally. Our results show that in many practical scenarios this self-adaptive scheme, which does not require dedicated resources for propagating of load information and decision making, performs extremely well with respect to best known practice.

Link dimensioning is generally considered as an effective and (operationally) simple mechanism to meet (given) performance requirements. In practice, the required link capacity is often estimated by rules of thumb, such as  =  ·, where is the (envisaged) average traffic rate, and some (empirically determined) constant larger than 1. This paper studies the viability of this class of ‘simplistic’ dimensioning rules. Throughout, the performance criterion imposed is that the fraction of intervals of length in which the input exceeds the available output capacity (i.e., ·) should not exceed , for given and .

We first present a dimensioning formula that expresses the required link capacity as a function of and a variance term (), which captures the burstiness on timescale . We explain how and () can be estimated with low measurement effort. The dimensioning formula is then used to validate dimensioning rules of the type  =  ·. Our main findings are: (i) the factor is strongly affected by the nature of the traffic, the level of aggregation, and the network infrastructure; if these conditions are more or less constant, one could empirically determine ; (ii) we can explicitly characterize how is affected by the ‘performance parameters’, i.e., and .

We investigate the problem of ensuring and maximizing performance guarantees for applications suffering software aging. Our focus is the optimization of the minimum and average performance of such applications in virtualized and non-virtualized scenario. The key technique is to use a set of simultaneously active application replica and to optimize their rejuvenation schedules. We derive an analytical method for maximizing the minimum “any-time” performance for certain cases and propose a heuristic method for maximization of minimum and average performance for all others. To evaluate our method we perform extensive studies on two applications: aging profiles of Apache Axis 1.3 and the aging data of the TPC-W benchmark instrumented with a memory leak injector. The results show that our approach is a practical way to ensure uninterrupted availability and optimize performance for even strongly aging applications.

The discovery of dependencies between components of a network can reveal relationships among components of multi-tier applications and the underlying IT infrastructure, such as servers and databases. Knowledge of these dependencies is thus important for the management of large distributed, heterogeneous and virtualized systems, where it is difficult to maintain an accurate view of how network assets are functionally connected. In this paper we present a passive method that uses attributes of traffic flow records and derives traffic dependencies among network components using a flexible fuzzy inference mechanism. Simulations and evaluation with real traffic traces show the applicability of the approach for flow-based dependency detection.