Work in Progress: Automating Proportion/Period Scheduling

The recent effort to define middleware capable of supporting real-time applications creates the opportunity to raise the level of abstraction presented to the programmer. We propose that proportion/period is a better abstraction for specifying resource needs and allocation than priorities. We are currently investigating techniques to address some issues that are restricting use of proportion/ period scheduling to research real-time prototypes. In particular, we are investigating techniques to automate the task of selecting proportion and period, and that allow proportion/period to incorporate job importance under overload conditions. Priorities, the abstraction that is currently the most widely deployed in both real-time and non-real-time systems, suffer from two key drawbacks. First, as a means of requesting resources, priorities can only express two forms of resource sharing: all-or-nothing and equal share. In particular, one cannot request proportional sharing or express timing or smoothness requirements using priorities. Second, priorities provide no means of isolating one application from another. As a result, one can only model or predict the behavior of an application by knowing the behavior of every other application that is running, or by running one’s application at a higher priority than every other application on the system. The former technique is impractical on large scale and distributed systems, while the latter technique only works for one application. A common alternative to priorities is to express resource needs in terms of proportion and period. Under this scheme, jobs request a percentage or proportion of the resource every period time units. This scheme solves the problem of expressibility, by giving the ability to express proportional share as well as smoothness requirements. One can achieve proportional sharing by allocating different proportions to jobs, such as by giving job A 60% of the resource and job B 40%. One can minimize burstiness by specifying a small period. This scheme also isolates one job’s resource allocation from another’s. This allows programmers to reason about independent applications independently. In addition, proportion/period schemes can detect and respond to job starvation before it arises, allowing schedulers to perform admission testing or job cancellation. Despite these advantages, proportion/period scheduling has yet to see main stream appeal due to two impediments. First, current schemes only support reservations of resources, which force application programmers to determine worst-case behaviors a priori.[2][3][5] The use of reservations forces unacceptable development costs to determine these behaviors accurately, particularly for highly dynamic applications and environments. Second, current schemes do not support the notion of job importance. Typically, jobs are admitted to the system (e.g., allowed to run) if the system can guarantee their resource request, and jobs that have been admitted are allowed to continue until they terminate. This scheme does not allow the arrival of an impor-