Systems
Two circuits, one trench
You are paying for a second internet connection so that the site keeps running when the first one fails. Whether it will depends on a fact neither invoice states: whether the two of them share anything.
In short
- Redundancy is a property of what two links do not share, not of there being two of them. Two circuits that share a conduit, a building entry, a local exchange or an upstream carrier fail together for the reason you bought the second one.
- Reselling is the version nobody suspects: two different companies will happily sell you two circuits that are the same physical line from the same underlying carrier, and neither bill will say so.
- Availability multiplies only across independent failures. Two links that are truly independent turn two nines into four; two links sharing one trench are one link with two invoices.
- A failover that has never been exercised is a claim rather than a capability. The common discovery during a real outage is that the switch works and something behind it, usually a phone system or a site-to-site tunnel, does not follow.
- Different technology is a reasonable proxy for different physical path when diversity cannot be confirmed. Fiber plus a cellular link is rarely elegant and rarely shares a trench.
A site that cannot operate offline buys a second circuit. It is the right instinct and it is often the first genuinely sensible infrastructure decision a growing company makes. The question that decides whether the money did anything is one that neither provider will volunteer and neither invoice contains: what do these two connections have in common?
Because redundancy is not a count. It is a statement about independence, and a second link only buys you something across the failures it does not share with the first. The useful version of this question is not whether you have two circuits. It is what single event takes both of them down, and there is nearly always one.
The four things a backup link commonly shares¶
| What is shared | The event that takes both down | How to find out |
|---|---|---|
| The building entry | Anything that happens to the one point where cabling enters the building, which is frequently a contractor with a digger | Walk it. This is the one you can settle yourself in twenty minutes with a torch. |
| The duct or trench to the street | The same excavation, and it is the most common cause of a genuinely diverse pair failing together | Ask each provider for a diverse path in writing, and ask specifically about the final stretch to your building |
| The local exchange or cabinet | A power failure or equipment fault at one facility both services terminate into | Ask which facility each circuit terminates at. Providers will usually answer this when asked directly. |
| The underlying carrier | An outage at a company you have no contract with and may never have heard of | Ask each provider whether they own the physical infrastructure or resell somebody else's, and whose |
The fourth row is the one that surprises people, and it is worth stating as plainly as possible: two circuits bought from two different companies, on two different contracts, at two different prices, can be the same physical line from the same underlying carrier. Nobody has misled you. You asked for a second circuit and you were sold one. The question that would have surfaced it was never asked, because it does not occur to a buyer that the answer could be yes.
What the arithmetic says, and what it assumes¶
The reason two links are worth paying for is that independent failures multiply. If a single circuit is unavailable one per cent of the time, two independent ones are both unavailable one time in ten thousand, and that difference between one per cent and one hundredth of a per cent is the entire product you are buying.
Every part of that depends on the word independent. If the two circuits share one trench, the shared cause is not multiplied at all: it is simply the failure rate of that trench, and you are paying twice for the availability of one thing. This is why the question is worth pursuing past the point of politeness. It is not pedantry about terminology. It is the difference between two nines and four, and it is decided by facts that are knowable before you sign.
The failover nobody has exercised¶
Having two independent circuits and moving traffic between them are separate problems, and the second is where we find most of the real defects. The switch itself is usually fine. What tends not to survive is everything that quietly assumed the first circuit's address.
- The phone system, which is the one that gets noticed first and is frequently the reason the outage becomes a customer-facing event rather than an internal one.
- The site-to-site tunnel to another office or to a cloud environment, configured against the primary address and with no second definition.
- Anything a third party allowlists by address, which includes a surprising number of payment, EDI and supplier integrations.
- Inbound anything. A backup circuit commonly carries outbound traffic perfectly well and cannot accept a connection, which is invisible until the day it matters.
All four are found the same way, by unplugging the primary circuit deliberately, at a time you choose, with everyone warned. An organization that has never done this does not know whether it has failover; it has a configuration that is believed to provide failover, and the first genuine test will be conducted by circumstance at the least convenient hour available.
The questions worth asking before you sign¶
- 1.Do you own the physical infrastructure to this building, or are you reselling? If reselling, whose network is it?
- 2.Which facility does this circuit terminate at, and is it the same one as my existing service?
- 3.Will the final stretch into the building follow the same route as my existing circuit, and can you state that in the contract rather than on a call?
- 4.What is the committed restoration time when it fails, and what happens if you miss it? A number with no consequence attached is a hope.
- 5.When we test failover by disconnecting the primary, what should we expect to stop working? A provider who can answer that specifically is one who has seen it done.
None of these is a hostile question and a competent provider answers all five readily. The value is not in catching anyone out. It is that you will have written down, before the outage, what you actually bought, which is the document nobody has on the morning it is needed.
Next step
Send us both circuit bills.
The two invoices, with the circuit identifiers on them. We will tell you what we can establish about whether you are buying one path twice, and give you the specific questions to put to each provider to settle the parts a bill cannot answer.
- Phone
- (214) 723-2510
- Reply
- A person replies, not a sequence: within one business day, from someone who would be on the engagement.
Related reading
- Systems9 min read
What an estate review actually inspects, in order
Almost nobody replacing an IT provider has a current inventory, because the outgoing provider holds it. This is the method for building one from the estate rather than from an interview, and what each pass usually turns up.
- Delivery8 min read
The handover pack, published before you need it
Every managed service agreement promises reasonable cooperation on the way out. Nobody says what the artifact is. Here is ours, section by section, so you can hold it against whatever your current provider has agreed to.
- Security6 min read
Your plant network is flat, and the drawing says otherwise
Every site we have assessed had a network diagram showing a segmented plant. On most of them the segmentation was a VLAN with a permissive rule, a vendor laptop, or a spreadsheet nobody had opened in three years.