ALLOCATING RESOURCES FOR SOFT REQUESTS - A PERFORMANCE STUDY

To effectively utilize a Limited number of resources shared by many us ers, algorithms have been proposed to manage the ''dynamic resource al location'' problem. Traditionally, these algorithms only deal with rig id requests that precisely specify the desired number of resources nee ded. However, in applications such as storage allocation and parallel processing, or in new telecommunication services such as Switched Frac tional-DS1 and adaptable-bit-rate video, a user may prefer to initiate soft requests, each of which expresses not only the desired number of resources but also the minimum number of them. In order to increase t he possibility of a request being granted, the user who sends in the s oft request is willing to accept a decrease in service quality by rece iving fewer resources. I-few to decide whether to accept or deny soft requests, so that a performance objective can be met, adds a new dimen sion to the old resource allocation problem. This paper studies the im pact of soft requests on various resource allocation algorithms. Four allocation algorithms for soft requests are studied in this paper via a simulation that is based upon an event-driven model. For the perform ance measures we choose (i.e., utilization, acceptance ratio, and allo cation ratio), the results indicate that some algorithms can gracefull y adapt to different system loads and achieve both a high acceptance r atio and a high utilization at the cost of the allocation ratio. The r esults also show that having soft requests is better for enhancing the acceptance ratio, and that increasing the time-out period can further improve the possibility of a soft request being accepted. However, ou r study indicates that making the time-out period arbitrarily large ma y not be cost-effective, since after the time-out period is over a thr eshold value, the acceptance ratio is hardly improved further.