The development of Quality of Service (QoS) architectures for the Internet is a problem that has received significative attention recently. The large scale deployment of such architectures has being blocked by several reasons, such as, the dificulty in establishing Service Level Agreements (SLA) of QoS policies between users, Internet Service Providers (ISP) and operators, and the lack of mechanisms on the network to request services with quality guarantees. Despite of not having access to a service that can guarantee certain levels of performance per aplication, the Internet users are getting more and more concerned about the network quality. That can be confirmed by the number of users that are migrating to the broadband Internet access. The ISP´s, specially the broadband ones, must take measurements to manage their resources to satisfy the users´ expectations of quality and stay competitive on the market. In this work, problems have being identified at the processes used by a broadband ISP, the MLS Wireless. With the old system, certain situations could cause the denial of service due the number of simultaneos connections or degradation and increased latency due the lack of fairness to share the network resources. This work suggests some solutions for these deficiencies, implementing and analyzing their effectiveness. To fix the denial of service, we suggest the limitation of resources per user, so that the total capacity won´t be depleted. To have the resources more fairly distributed among the users, a new mechanism of sharing is implemented, based on a discipline named SFQ (Stochastic Fairness Queueing). The results show that those modifications have being effective. The time to access services like WWW and e-mail during congestion periods was drasticaly reduced, in some cases, from about 1 minute to less than 10 seconds.