Service-capacity allocation in an open migration process (source code)

= Service-capacity allocation in an open migration process
{title2=$\phi_j-\alpha_j\propto\sqrt{\alpha_j}$}

Consider an <open migration process> whose colonies are <M M 1 queues> with fixed positive traffic rates $\alpha_j$ and service rates $\mu_j>\alpha_j$. Their mean <queue populations> are $\alpha_j/(\mu_j-\alpha_j)$. Assume equal unit costs for capacity and a total rate budget $\sum_j\mu_j=F>\sum_j\alpha_j$.

Write $s_j=\mu_j-\alpha_j$ and $S=F-\sum_j\alpha_j$. The <Cauchy-Schwarz inequality> gives
$$
\left(\sum_j\sqrt{\alpha_j}\right)^2
\leq\left(\sum_j\frac{\alpha_j}{s_j}\right)\left(\sum_j s_j\right)
=S\sum_j\frac{\alpha_j}{s_j}.
$$
Equality holds exactly when $s_j$ is proportional to $\sqrt{\alpha_j}$. Thus the unique optimum is
$$
\mu_j=\alpha_j+\frac{S\sqrt{\alpha_j}}{\sum_i\sqrt{\alpha_i}}.
$$
For traffic rates $(1,4)$ jobs per second and budget $F=8$ jobs per second, the spare rates are $(1,2)$ and service rates $(2,6)$. The total mean population is $1/1+4/2=3$. Equal spare rates $(1.5,1.5)$ instead give $10/3$. The optimization allocates spare capacity rather than total capacity in square-root proportions; unequal capacity costs require a modified constraint.