In recent years, there has been growing momentum for studies on a sense of economy and marketability in computer systems. Taking account of users' satisfaction with resource allocation is addressed as a key component in economic approach such as game-theory, and also considered agents' incentives in theoretical models of artificial ecosystems. Discussion about resource contention on such environments can be applied to the notion of {\it open system} where users are potentially selfish and systems inherently involve uncertainty stems from their embodiment.
We employ a single shared medium that is associated with {\it pricing} but technically performs like the Ethernet as a test-bed for modeling an artificial ecosystem in terms of resource contention among selfish agents. In our model, each agent has a {\it gene} which characterizes its service strategy, and an {\it utility function} determines its attitude toward resource contention. Agents make efforts eventually to maximize own utility by selecting one agent as a contractor after another through trial-and-error.
We design and implement a network simulator based on our model, and conduct some simulations to observe selfish agents' behavior and look into dynamism of evolutional artificial ecosystems. Simulation results say that although there is no premise that agents always act in a cooperative manner, cost-sensitive agents can collectively self-organize into a mass which exhibits cooperative appearance at relatively low cost in the short term, and consequently, a network system can reach stable equilibrium as a whole.
In recent years, it has come to light that connection less internetworking protocols are inadequate for real-time continuous data delivery in terms of congestion control. For efficient congestion control, several connection-oriented internetworking protocols have been developed. These protocols generally depend upon resource reservation to avoid congestion, however, reservation inherently wastes network resources. Besides, we cannot use such protocols on an internet which contains sub-networks that do not support resource reservation.
We propose a new design of a connection-oriented internetworking protocol which we call ``CIP'' (Connection-oriented Internet Protocol). CIP has two major characteristics:
We present the implementation of CIP on Mach 3.0 as an application executed in the user address space, and evaluate it, focusing on efficient utilization of existing network resources.
In the face of recent progress found in increase of network bandwidth, overhead of CIP hop-by-hop flow control can become burdensome in the future. So we also incorporate mechanisms for rate control and buffer space pre-allocation into CIP connection management as provisions for future extensions of CIP.