Techniques are presented for automatic tuning of operating parameters, e.g., amplifier gain, in an optical network. A section of an optical network comprises a plurality of spans between optical nodes, and each optical node has an amplifier to amplify optical signals for transmission between optical nodes. Physical network layer data is obtained from the optical nodes for use as input to an analytical model. A set of powers defining an optimum working point of the amplifiers is computed based on variations in amplifier noise figure which depend on amplifier gain. A figure of merit representative of network section performance is computed based on linear and non-linear noise at current power levels of the amplifiers. The figure of merit is evaluated. The set of powers is applied to the amplifiers in the network section when evaluation of the figure of merit indicates that network performance improvement can be achieved.