Friday, 4:47 p.m.
The office does not go silent when the internet fails. It gets louder. Phones ring. A payment terminal spins. The shared drive disappears. Someone opens a hotspot and discovers the building has poor cellular coverage. At that moment, nobody cares what the speed-test screenshot looked like on installation day. They care whether orders can close, customers can reach the team, cameras are recording, and someone who understands the circuit is already working the problem.
The network stops being a line item and becomes what it was all along: a dependency shared by nearly every part of the business. The value of a network is not the speed it reaches when everything is healthy. It is the work the business can keep doing when something is not.
Bandwidth is the easy number
Providers lead with bandwidth because it is easy to print and compare. The number matters, especially for cloud backups, media, dense wireless environments, and many simultaneous users, but it is one dimension of the service. Two one-gigabit circuits can behave like completely different products.
Before choosing speed, translate the business into traffic. Which applications are interactive? Which transfers are large? What must continue during backup operation? Are voice, security, guest Wi-Fi, and production systems isolated? A circuit sized without those questions is a number searching for an architecture.
- Ask whether capacity is dedicated, committed, or best-effort.
- Identify applications whose performance changes revenue, safety, or customer experience.
- Measure ordinary and peak demand.
- Make sure the firewall, switching, and wireless layers can deliver what is purchased.
The path matters more than the logo on the invoice
A network is physical before it is digital. Glass runs through conduit, cable follows poles, or a radio establishes line of sight. The path enters the property, reaches a demarcation point, passes through powered equipment, and lands on the firewall. Every segment creates a failure domain.
Buying two services does not automatically create diversity. Different provider names can share a pole, conduit, manhole, building entrance, riser, upstream carrier, or utility power. If both services cross the same single point, they may fail together.
Detroit makes the physical conversation especially important. Buildings span more than a century of construction. Some have reusable pathways; others have crowded risers, undocumented renovations, or one practical street entrance. The clean line on a map is not the same as a surveyed, permitted, constructible route.
Diversity is not two cables. It is the deliberate removal of a shared failure domain.
A second circuit needs a job description
Backup connectivity often begins as a purchase instead of a design. Give the secondary path a job. It may need to preserve every application, or it may carry only payments, voice, identity, and remote access while backups and guest Wi-Fi pause. The medium is a design choice; independence and operating behavior are the goals.
Define the failure signal, essential traffic, inbound and outbound behavior, voice and VPN behavior, and the return to primary. Complexity is valuable only when the team can operate and test it.
Power is part of the circuit
A fiber strand does not need electricity, but nearly everything that makes it useful does. Trace every active device between the carrier handoff and the user. Decide which devices require uninterruptible power, how long batteries must carry them, whether generator power is available, and how cooling is handled. For a critical site, power monitoring belongs beside link monitoring.
An SLA is an operating agreement
A useful service-level agreement defines how availability is measured, what is excluded, when the clock begins, what response and restoration targets apply, how maintenance is communicated, and what happens when the commitment is missed. A credit can create accountability, but it rarely compensates for the real cost of an outage. The customer still needs a continuity plan.
Read the support model carefully. Can the first team see the circuit and route? Can they separate building power from a carrier-path fault? Is there a documented route to someone who can change something?
Monitoring should create action, not wallpaper
A graph becomes operational when somebody knows what a change means, when to act, and who owns the next step. Monitor from more than one perspective: interface state, route health, latency, loss, equipment health, and application checks reveal different failures. Good alerts include context and lead to an action decided in advance.
Test the recovery path on a calm day
A backup connection that has never carried production traffic is a theory. Schedule controlled tests. Define success, capture the starting state, fail the primary path, observe what users and applications experience, and measure detection, transition, degradation, and recovery. Then update the configuration and runbook while the lessons are fresh.
The conversation to have before you buy
A strong provider conversation should feel less like choosing from a menu and more like reviewing a system. Bring the business impact, building reality, applications, diagram, and failures that keep you up at night. You should leave knowing what is committed, estimated, shared, and still your responsibility.
At Octet, this is the standard we want applied to us. We are building in Detroit because local infrastructure deserves local attention: people who can look at the route, understand the building, connect the service to the business, and remain accountable after the install photo is taken.
- What workflows must the network protect?
- Where are the physical and upstream failure domains?
- What is monitored, by whom, and what action follows?
- How will failover behave, and when was it tested?
- Who answers when the ordinary path fails at 4:47 p.m.?