Building distributed systems according to the Service-Oriented Architecture (SOA) allows simplifying the integration process, reducing development costs and increasing scalability, interoperability and openness. SOA endorses the reusability of existing services and aggregating them into new service layers for future recycling. At the same time, the complexity of large service-oriented systems negatively reflects on their behavior in terms of the exhibited Quality of Service (QoS). To address this problem I propose to employ request scheduling as means for meeting Service Level Agreements (SLAs). Special focus is given to composite services specified using workflow languages. The proposed solution suggests exercising two-level scheduling: global and local. The global policies assign the response time requirements for component service invocations. The local scheduling policies are responsible for performing request scheduling in order to meet these requirements.  In this talk, I will present the proposed approach and discuss the results of its empirical evaluation.