Current Hard Real Time (HRT) kernels have their timely behaviour guaranteed on the cost of a rather restrictive use of the available resources. This makes HRT scheduling techniques inadequate for use in Soft Real Time (SRT) environment where we can make a considerable profit by a better and more flexible use of the resources. We will show that we can also improve the responsiveness and fairness of SRT kernels by combining Real Time Transactions and the Earliest Deadline First rule with Deadline Inheritance policies. In particular we present a policy where each task can request its own private Scheduling Quality of Service from the kernel. The kernel can offer scheduling services with varying levels of service ranging from "not attentive" to "very attentive". We claim that the proposed techniques (1) can be combined with HRT techniques such that HRT guarantees are not affected, (2) that they are particularly useful to multimedia environments, (3) that they are easy to use, and (4) that the required administration overhead is low. A currntly running multimedia experiment shows a scheduling overhead of less than one percent.
|Number of pages||17|
|Publication status||Published - 1996|