The current solutions for the parallel execution of scientific workflows are appropriate for static computing and storage resources in a grid environment. They have been extended to deal with more elastic resources in a cloud, but with only one site. Our analysis [1] of the current techniques of scientific workflow parallelization and scientific workflow execution has shown that there is a lot of room for improvement in the following directions: 1. Data staging: existing techniques mainly focus on the mechanism that starts scientific workflow execution after gathering all the related data in a shared-disk file system at one data center, which is time consuming. 2. Architecture: the structure of SWfMSs is generally centralized, with a master node, which is a single point of failure and performance bottleneck, managing all the optimization and scheduling processes. 3. Task scheduling and data location: most SWfMSs do not take data location into account during task scheduling, which makes it inefficient to read or write data. 4. Multisite: novel task and data scheduling approaches are required for utilizing resources in a multisite cloud. In the rest of this paper, we define more precisely the problem and introduce our approach to address it.