Reifying Control of Multi-Owned Network Resources

Abstract

Communication delay is a key source of uncertainty in distributed systems. Existing approaches to reduce this uncertainty focus on maintaining sufficient surplus bandwidth; applications, on their part, are designed in ways to tolerate certain degree of uncertainty in communication delays. This leads to contention between the goals of optimal utilization and acceptable delays. We argue that the multi-owned nature of today's networks offers opportunities to reason about and scalably control networks at a fine grain. An explicit treatment of network resource ownership and trade allows reasoning about acceptable delays. This can lead to scalable mechanisms for fine-grained accounting and reification of control, which make it possible to quantify and control network utilization. We bring together ownership, fine-grained accounting, and reification of control in a model for resource acquisition and control called CyberOrgs. CyberOrgs encapsulate distributed computations with resources required for their execution. A CyberOrg acquires resources required by its computations by buying them from other cyberorgs using eCash. We present a novel approach for implementing finegrained network resource control based on the CyberOrgs model. A prototype implementation is described with experimental results illustrating the effectiveness of control.