Many scientific application which requires large volume of data processing can benefit from the evolution of the cloud computing which enables users to get utility computing services. However, as there exists many kinds of services offered to the cloud users in various types of resources and contracts, users who want to process their workflow execution request reasonably may feel difficult to utilize those resources by themselves. Therefore, the cloud workflow service broker is introduced to define the scientific workflow execution request then orchestrate resources to process submitted request. The goal of the broker is providing services which satisfy user-specified Quality of Service (QoS) constraints such as time and cost. Accordingly, the Service Level Agreement (SLA) is made between the user and the broker to define the region where user-specified QoS constraints can be satisfied. However, SLA violation might happen due to unexpected resource performance variance. Also, when user re-quires to process the workflow request within tough or insufficient QoS constraints, the cloud broker should con-sider rejecting the request. Therefore, workflow scheduling based on user-specified QoS constraints issue arises in the broker. In this thesis, we propose the QoS constraints workflow scheduling scheme utilizing the colored Pe-trinet model to apply the task division policy which enables to expand the QoS-guaranteed range, and effective-ly cope with resource performance variance. The proposed algorithm investigates each task’s workload then decides its distribution rate. Afterwards, the proposed algorithm allocates the cheapest VM to each task which can satisfy the subdeadline of the task with respect to the longest path on workflow topology described by Petrinet model. If there is no suitable VM resource, the task division policy is applied while penalty cost is considered. We compared the performance of the proposed algorithm, which is called as the Phased Workflo...