Central Agent Registry. Isolated Requests 69 5.3.2 Central Agent Registry: Sessions 73 5.3.3 Central Service Registry 76 5.4 Conclusion 77 6 Decentralised Service Location 78 6.1 Introduction 78 6.2 Organisational Structures 79 6.3 Network Based Search 82
Central Agent Registry. Isolated Requests The problem of locating services can be simply solved by using a central agent registry that maintains contact data on all agents in the system; it can then be queried to find a single random agent. If an agent k needs to find another agent offering service s, k first sends a request to the registry, and the registry replies with details of a random agent m1. Agent k then asks m1 if it offers service s, and m1 replies with confirmation if it does, at which point k allocates the task to m1. If it does not offer the service, then it replies with a denial, at which point k requests from the registry another agent’s details, m2. This centralised search is very simple, and ultimately does not fail to locate services (assuming no communication problems). However, since the registry does not track requests, it is possible that m1 = m2, and so multiple denied requests to the same agent are also possible. Because the registry does not consider the result of previous requests when responding to a query, we say that this approach uses isolated requests.
Central Agent Registry. Isolated Requests Our first experiment evaluated task completion time when locating services by making isolated requests to a central agent registry. For this approach, an average service location time of 400 time steps is achieved, as shown in Figure 5.1, but service location can — on rare occasions — take up to nearly ten times this. This is shown by the frequency distribution of service location time in Figure 5.2(a), which highlights a tail-off effect such that services are generally located quickly, but on rare occasions service location takes a significantly larger amount of time. The bars in the graph indicate the number of instances for which service location took a value in a limited range. The graph is shown as a histogram for ease of inspection, and the results can be understood probabilistically, as follows. | Suppose we have a set of agents a, only one of which offers the required service. A request is sent to the central registry, and the identity of an agent is returned. Now, the probability of the returned agent offering the required service is 1 , and conversely,
Central Agent Registry. Sessions Our second experiment evaluates task completion time when locating services through a central agent registry that maintains sessions for each service location process. By remembering the agents that have already been contacted during a particular search process the tail-off effect, present in the previous service location approach, can be removed. In fact, in doing so, this makes service location time much more predictable. The time required to check if one agent offers a particular service is four time steps (a request and reply to the central registry, plus a request and reply to the agent identified). Thus, for a system of 100 agents, the time to locate a service is always somewhere between 0 and 400 time steps, as shown in Figure 5.2(b), and the average time needed to locate a service is 200 time steps, as shown in Figure 5.1. More generally, for a system that consists of n agents, service location time is, in the worst case, 4n and, on average, 2n. This performs better than isolated requests, but requires the registry to keep a ● ●● ● ●●●● ●●● ● ●● ●● ●● ●● ●●● ●● ●●●●● ●●●●● ● ● ●●●●●●●●● ●● ●●● ●● ● ● ●● ● ●●●●● ●●● ●●●● ● ● ● ●●●●● ● ● ● ●●●●● ●●●●●● ● ● ●● ●●●●●●●●●●● ●●●●● ● ●●●● ● ● ●●● ●●● ● ● ●●●●● ● ● ●● ●●● ● ●●●●● ● ●●●●●● ●●●● ●● ●● ●●●● ● ●● ●● ●●● ● ● ●●●●● ● ●●● ● ●●●● ● ●●● ● ●●●● ●●● ●● ●●● ● ● ● ●● ●●●● ●●●● ● ● ●●● ●●●●● ● ● ● ●● ● ●●● ●● ●● ●● ● ●●●● ● ●●●● ● ● ● ● ●● ●●● ●● ●● ● ●●● ●● ●●● ● ●●●● ●● ●●● ●●●●●●●●●●●●●●● ● ●●●● ●●● ●● ● ●● ●●● ●● ●●●●●●●●●● ● ●●●●●● ●● ●●●●●●● ●●●●●●●●● ●●● ● ●●●●●●● ●●●●●● ● ● ●●● ● ●●●●●● ●●● ●● ● ● ● ● ●● ● ●●●●●● ● ●● ●●●●●●●●●● ●● ●● ●●●● ● ● ●● ●● ● ● ●●●●●●●●●●● ● ● ●●●● ●●● ● ●● ●●●● ●●●● ● ● ●●●● ● ●●● ● ● ●● ●●● ●●●● ●●●● ●●●●●●● ●●●●●●●●●● ●● ● ●●● ●●●● ●●●●●● ●● ●●●● ●● ●●● ●● ●●●●●●●●●●● ● ●●●●● ●● ●● ●●●● ●●● ● ●●●● ●●●●● ●●●●● ●●●●●●●●● ●● ●● ●● ●●●●●●●●● ●●●●●●●●●●●● ● ●●●●●●●●●●● ●●●●●●● ●● ●●●●●●●●● ●●●●●● ●●●●●●●● ●●●●●●●●●●●●●●●●●● ●●● ● ●●●●●●●●●●●●●●● ●●●●●●●●●●●● ●●●●●●●● ●●●● ●●● ●●●●● ●●●●●●●●●●●●●●●● ●●●●●● ● ●● ●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●● ●●●●● ●●●● ●●●●●● ●● ●● ●●●●●● ● ●●● ●●●●●●● ●●●●●●●●●●● ●●●●● ● ●●●●●●●●●● ●●●●●● ●●●●●● ● ●●● ●●● ● ● ●● ●●●●●●● ●●●●●● ●● ●● ●●●●●●●●●●●●●●●●● ●● ●●●●●●●● ●●●● ●●●● ● ●●●●●●● ● ●●●● ●●●● ●●●●●●●●●●●●● ●●●●●●●● ●●● ●●● ●●●●●●●● ●●●● ● ● ●●● ●●●●●●●● ●●●●●●●●●●●●●●●● ●●● ●●● ●● ●●●●● ●●● ●●●● ●●●●●● ● ●● ●●●●● ● ●● ●●●●● ●●● ●●●●●● ●●●● ● ●●●● ●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●● ●●● ● ●●●●●●● ●●●●●● ●●● ●●● ●●●●●●●● ...
