Abstract

With the recent development of kernel patches for implementing kernel preemptability and high-resolution timers (and the acceptance of the first patch in the 2.5 development kernels), supporting real-time applications in Linux user-space is no more a dream. However, real-time applications must be run using the POSIX fixed priority scheduler, and we believe that this solution presents… a number of problems: for example, either a non-root user is not enabled to run real-time applications (hence, real-time support is useless for most users), or it has the power to starve all the other users' processes (even the daemons, and the root's processes). It is our opinion that to safely support real-time applications a multi-user OS such as Linux should provide Resource Reservations to enable users to access the real-time facilities without being able to starve the system. Resource Reservations are not a new concept, and they have been implemented in some Linux variants (the most notable is Linux/RK). In this work, we propose a new implementation, that fundamentally differs from the previous ones in some important points: 1) Our implementation is non-intrusive, i.e, the whole reservation scheduler is a kernel module that can be loaded at run time, and requires only few changes to the kernel; 2) Since the modifications to the Linux kernel are very limited, we can easily upgrade to new kernel versions, or take advantage of new features implemented in Linux, such as kernel preemptability and high-resolution timers; 3) The scheduling infrastructure that we use is very general, and can be used to implement other loadable schedulers; 4) Compatibility with standard Linux is maintained; 5) Thanks to the algorithm that we implemented (the Constant Bandwidth Server – CBS) and to the careful design of the scheduler infrastructure, our scheduler can cope with aperiodic task arrivals (and with tasks that block and unblock), that cause problems in other reservation-based systems. In this paper, we present the design of our scheduler (showing also how to implement different schedulers based on the scheduling infrastructure that we propose), we compare it with some previous works, and we present a preliminary performance and overhead evaluation. Finally, we describe how a user-level or kernel-level QoS manager can be implemented based on our scheduler, and can be used to enforce a CPU allocation policy, to avoid that a single user can reserve too much CPU to its real-time applications.

Read full paper (PDF download)

 
This article was originally published on LinuxDevices.com and has been donated to the open source community by QuinStreet Inc. Please visit LinuxToday.com for up-to-date news and articles about Linux and open source.