Infrastructure built so it would not be built again

The problem

The company had grown and the infrastructure had not. Nobody decided to end up in that situation: it was reached by adding patches, each reasonable at the time, until the whole thing lost its shape.

In the datacenter, critical services were running on desktop computers instead of servers. Old cabling with no structure, disorganised racks, network equipment long past its time. And connections and devices sitting there, working, with nobody able to say what they were for.

On the logical side the picture was the same: IP addressing assigned by hand over public ranges, no real segmentation between corporate, industrial, management and guest traffic. Documentation of applications and services was not reliable, so knowing how anything worked lived in a few people’s heads.

That last part is what stops a company from growing: every new site depends on one particular person being available.

The decision

Before touching anything, a full survey: what was there, what was obsolete, where the single points of failure were. Out of that came a transformation plan, and with it the two decisions that matter.

First: not to execute all of it. Part of the plan was implemented, and part was handed over as a proposal with support so the client’s own team could carry it out. It is a decision suppliers do not usually take, because every stretch the client executes is a stretch that does not get invoiced. It was taken anyway, because in the areas where the internal team had the capability, doing it themselves left behind something no external implementation leaves: people who know why their infrastructure is the way it is.

Second: not to solve only this site. Alongside the redesign, a replicable infrastructure and support model was defined, meant as the standard for the sites the organisation would open later. Instead of leaving a solution, leaving the template for the solution — so the next site would not require buying the same diagnosis again.

The result

The site ended up with the network redesigned end to end: new addressing, segmentation between corporate, industrial, management and guest networks, standardised equipment, perimeter security, the WAN between sites designed, the cabling redone and the racks built and ordered. With high-availability mechanisms where there had been a single point of failure.

Server virtualisation was also supported, along with migrating what had been running on machines never meant for it, and image-level backups.

But the deliverable that lasts longest is the other one. The infrastructure was documented, with a defined support and operating model, and with a replicable template for the sites that followed. The organisation stopped depending on the informal knowledge of a few people — which was the real limit on its growth, more than any ageing equipment.