Numerical efficiency of PESTO

Context

This short note aims at fixing the ideas about magnitudes of performance estimation problems we can expect to solve on a regular computer, with PESTO. We ran our tests on a local server with the following characteristics:

• lCPU Xeon W-2295 CPU @ 3GHz, 18 cores/36 threads ; RAM 128 Go ; Ubuntu 20.04 LTS
• 1, 2 or 4 threads MOSEK depending on the cases (this is a shared computer), baseline MATLAB memory approx. 1Gb.

Numerical results

For with "H_FastGradientMethod.m" (in the examples) and different values of N:

N = 50 : totaltime: 35 pestotime: 30 yalmiptime: 1 solvertime: 4 ; max memory 1.3 Go?

N = 100 : totaltime: 195 pestotime: 153 yalmiptime: 2 solvertime: 39 ; max memory 1.7 Go

N = 150 : totaltime: 786 pestotime: 451 yalmiptime: 4 solvertime: 330 ; max memory 3.6 Go

N = 200 : totaltime: 2210 pestotime: 1042 yalmiptime: 7 solvertime: 1157 ; max memory 7.0 Go

N = 250 : totaltime: 7014 pestotime: 1948 yalmiptime: 12 solvertime: 5046 ; max memory 15.1 Go

N = 300 : totaltime: 17401 pestotime: 3182 yalmiptime: 18 solvertime: 14185 ; max memory 27.8 Go