We propose a novel distributed approach to exploit sleep mode capabilities of links in an Internet Service Provider network. Differently from other works, neither a central controller, nor the knowledge of the current traffic matrix is assumed, favoring a major step towards making sleep mode enabled networks practical in the current Internet architecture. Our algorithms are able to automatically adapt the state of network links to the actual traffic in the network. Moreover, the required input parameters are intuitive and easy to set. Extensive simulations that consider a real network and traffic demand prove that our algorithms are able to follow the daily variation of traffic, reducing energy consumption up to 70% during off peak time, with little overheads and while guaranteeing Quality of Service constraints.