The primary goals of self-management are twofold: cost reduction and automated adaptive control of complex network operations. The optimal allocation of resources in dynamic networks has emerged as a recurrent denominator for a branch of self-optimization problems. Like most algorithms for self-management, traditional optimization algorithms are challenged by specific aspects of networks: complexity, scalability, and time variability. This paper proposes to address this class of problems with robust, adaptive, and distributed optimization methods, in order to cope with high levels of dynamicity and uncertainty, while still preserving good convergence properties. We illustrate the benefit of using such methods through two different problems occurring at different network layers: 1) at the transport layer, the flow admission control driven by quality of service (QoS) requirements; 2) at the service layer, the negotiation of service level agreements to sustain multi-domain QoS guarantees.