Motivation

One of my favorite Procurement SYSTEMS¹ is the Federal Procurement System of USA, being the GSA (General Services Administration²) its main component.

---

Here I explore data from the extensive open data catalog of GSA. What captured my attention this time, was data from several Data Center closings due to a new vision about the benefits of centralization in these services. FDCCI is the acronym for Federal Data Center Consolidation Initiative, which aims agencies “[…] to optimize and consolidate data centers to deliver better services to the public while increasing return-on-investment to taxpayers.”³

---

How important can be this initiative? According to data published in ITDashboard⁴ only during 2013-2015 (just part of the period considered here!), the LIFECYCLE COST of investment in the datacenter item is $USD 4.503.810.021.

---

I’m skeptical of centralization of these type of services, for a very simple reason: Let \(var\) be the variance of a random variable and \(cov\) its covariance, then \(var(x_1 + x_2) = var(x_1)+var(x_2) + 2 cov(x_1 , x_2)\). Let \(\sigma_1 ^2\) be equal to the variance of \(x_1\) and \(x_2\), then \(cov(x_1 , x_2 ) = p\sigma_1 \sigma_2 = p\sigma_1^2\), where \(p\) is correlation coefficient; hence, our variance expression becomes: \[var(x_1 + x_2) = 2(1 + p)\sigma_1^2\] - If only one data center is contructed/operated for every organization (let’s call this full centralization), then \(x_1 = x_2\) and \(p = 1\) then \[var(x_1 + x_2) = 4\sigma_1^2\] - If the data centers for each organization are independent, then \(p = 0\) and \[var(x_1 + x_2) =2\sigma_1^2\] - And also interesting, if the data centers are inversely correlated (here we talk about security and availability), then \[var(x_1 + x_2) < 2\sigma_1^2\]

So clearly, centralization is the worst response, in case of cyber security problems, but also, in terms of the risks incurred during contracting.