Design And Implementation Of Real-time Scheduler In Xen For Mixed-Criticality Tasks On Multi-core Platforms

System integration will be a significant trend for real-time embedded systems in the future.Although system integration brings great improvement in cost,size,weight and power efficiency,the real-time guarantees required by tasks make it to be a great challenge.There are 3 ways to integrate real-time embedded systems:(1)As to hardware,single-core platforms are substituted by multi-core platforms.For the reason that normally network is used for communication by multiple subsystems on single-core platforms,the limitation of bandwidth and transmission latency will bring complexity to scheduling real-time tasks and become unable to provide reliable real-time guarantees.(2)As to software,tasks of different criticalities are integrated on one shared hardware platform in order to reduce cost,weight and improve power efficiency.(3)Virtualization Technology is adopted.Virtualization Technology is used to provide efficient isolation for the virtual machines concurrently running on the shared platform to increase the utilization of hardware resources.In this thesis we try to provide support to all these techniques by designing a scheduling scheme in real-time systems.We tackle the problem of "providing support for scheduling real-time tasks and mixed-criticality tasks on multi-core platforms in virtual machines".To solve this we propose a scheduling scheme and implement a scheduler which is able to support all the three techniques in the open-source VMM Xen.The scheduling strategy of our CBS-MC scheduler is based on resource reservation.CBS-MC scheduler improves the algorithm of Constant Bandwidth Server(CBS)according to actual environment to provide support for mixed-criticality tasks.CBS-MC scheduler brings 3 promotions to real-times systems:(1)The predictability of the temporal behavior of real-time tasks in virtual machines is enhanced.(2)Reliable temporal isolation for virtual machines is provided.(3)The response times of low-critical tasks are reduced.In this thesis all these features are verified by intensive tests.Additionally,an experimental platform is implemented for testing and calculating response times of jobs.Instead of simulation,we create processes and threads which execute as real jobs to get more convincing results.The finishing time of jobs are gathered and processed into cumulative distribution function.