Name
Abstract | ||
|---|---|---|
The allocation of resources in networks is stud- ied in the context of the influence model, a stochastic net- work model composed of interacting Markov chains. First, some new results regarding the influence model are pre- sented. Next, a specific resource allocation problem is posed, and three different methods for allocating resource are described and analyzed. Both the structure of the re- source allocation and the resulting behavior of the network are explored for each method. Finally, some more general insights about these resource allocation procedures are dis- cussed. even random. Also, resource allocation may be static (de- termined once) or adaptive (altered with time); it may be locally controlled (determined by small parts ofthe net- work) or globally controlled (based on the decision ofa single controller); the resource may be allocated all at once or sequentially; the resource allocation may be constrained to be homogeneous (uniform throughout the network) or may be inhomogeneous (allowed to vary throughout the network); the resources may be allocated to the nodes in the network or the branches connecting these nodes; and many other categories for resource allocation procedures can be suggested. A thorough study ofall the possible resource allocation methods listed above would be a difficult task, even in the limited context ofthe influence model. In this article, we will describe three specific resource allocations that we be- lieve elucidate important features of real networks: a static homogeneous design based on a global cost function; a dy- namic inhomogeneous design based on a global cost func- tion; and a reactive, dynamic, and inhomogeneous resource allocation procedure. We will also compare these three resource allocation procedures to gain some more general insight into resource allocation methods. Some explanation for why the influence model is used in our work is important. We believe that the influence model captures many ofthe important features ofnetworked sys- tems, including the interdependence among nodes in the network and importance ofstochastic events to network dynamics. Furthermore, the quasi-linear structure ofthe model greatly enhances its tractability, so that we can an- alyze different resource allocations and understand the re- sulting behavior ofthe model. The remainder ofthe article is organized as f ollows: • Section II: We summarize the influence model and dis- tinguish between the homogeneous and inhomogeneous in- fluence model. • Section III: Some new analyses ofthe influence model are described. These results will subsequently be used for exploring resource allocation in the model. • Section IV: Three different resource allocations are de- scribed and analyzed in the context ofthe influence model. • Section V: We discuss the similarities and differences among the different resource allocation problems. Using these comparisons, we postulate important general char- acteristics ofresource allocation in networks. Conclusions are presented and directions for future work are briefly dis- |
Year | DOI | Venue |
|---|---|---|
2002 | 10.1109/HICSS.2002.993978 | HICSS |
Keywords | Field | DocType |
resource allocation,network dynamics,markov processes,markov chain,cost function | Markov process,Computer science,Markov chain,Resource allocation,Network model,Distributed computing | Conference |
Citations | PageRank | References |
3 | 0.55 | 1 |
Authors | ||
3 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| Sandip Roy | 1 | 301 | 53.03 |
| Bernard C. Lesieutre | 2 | 116 | 32.96 |
| George C. Verghese | 3 | 208 | 26.26 |