Bandwidth Reservation strategies

Due to the implicit temporal semantics of CM, a real-time support is useful to manage audio/video streams. However, the use of a hard real-time system for handling CM applications can be inappropriate for the following reasons:

• If a multimedia task manages compressed frames, the time for coding/decoding each frame can vary significantly, hence the worst case execution time (WCET) of the task can be much bigger than its mean execution time. Since hard real-time tasks are guaranteed based on their WCET (and not based on their mean execution times), CM applications can cause a waste of the CPU resource.
• Providing precise estimation of WCETs is very difficult even for those applications always running on the same hardware. This problem is even more critical for multimedia applications, which in general can run on a large number of different machines (think of a video conferencing system running on several different PC workstations).
• When data are received from an external device (for instance, a communication network) the interarrival time of the tasks processing such data may not be deterministic, so it may be impossible to determine a minimum interarrival time for such tasks. As a consequence, no a priori guarantee can be performed.
• Advanced multimedia systems tend to be more dynamic than classical real-time systems, so all the scheduling methodologies deviced for static real-time systems are not suited for CM applications.

The biggest problem in using hard real-time techniques to schedule multimedia taskx is that a task requiring too much computation time (overloaded task) can affect the QoS experimented by other tasks. It is instead desirable an isolation property, that ensures that any overload affects only the task that generated it.

This isolation property can be obtained through bandwidth reservation techniques: each task is reserved a fraction of the CPU bandwidth and it executes requiring no more than the assigned bandwidth. For example, if a task T1 is assigned 20% of the CPU bandwith, it will execute as if it were the only task on a CPU that executes at a speed that is 20% of the real CPU speed

In this way, if the sum of all the reserved bandwidth is less than 100%, each task will not be affected by other tasks' overloads.

We implemented a Bandwidth Reservation strategy in the HARTIK kernel to support multimedia computing: multimedia activities can be integrated with hard real-time, soft real-time and non real-time activities without jeopardizing hard real-time tasks' a priori guarantee.

Our Bandwidth Reservation strategy is realized serving each multimedia task with a dedicated Constant Bandwidth Server (CBS). The semantic of this server ensures that a task served by a CBS S1 never demands a bandwidth greater than S1 bandwidth.

Go to HARTIK and Multimedia

Go to HARTIK Home Page