Abstract:
The resource allocation problem in computer networks with a large number of links is considered. The links are used by consumers (users), whose number can also be very large. For the dual problem, numerical optimization methods are proposed, such as the fast gradient method, the stochastic projected subgradient method, the ellipsoid method, and the random gradient extrapolation method. A convergence rate estimate is obtained for each of the methods. Algorithms for distributed computation of steps in the considered methods as applied to computer networks are described. Special attention is given to the primal-dual property of the proposed algorithms.
Key words:
resource allocation, communication networks, network utility maximization, primal-dual property, fast gradient method, stochastic projected subgradient method, ellipsoid method, random gradient extrapolation method.
Gasnikov’s research was supported by the Russian Foundation for Basic Research, grant no. 18-31-20005 mol_a_ved and 19-31-51001 Scientific mentoring. Dvurechensky’s research was supported by the Russian Foundation for Basic Research, grant no. 18-29-03071 mk. Vorontsova’s research was supported by the Ministry of Science and Higher Education of the Russian Federation (state assignment no. 075-00337-20-03), project no. 0714-2020-0005.
Citation:
E. A. Vorontsova, A. V. Gasnikov, P. E. Dvurechenskii, A. S. Ivanova, D. A. Pasechnyuk, “Numerical methods for the resource allocation problem in a computer network”, Zh. Vychisl. Mat. Mat. Fiz., 61:2 (2021), 312–344; Comput. Math. Math. Phys., 61:2 (2021), 297–328