Algeria AgencyNetwork & connectivity
Networks: the cheapest layer, and the one that produces half the faults
A network is a chain: it is worth what its worst segment is worth. That is why it gets bought in the order the signal travels.
The article on the IT pillar says the network is the cheapest layer and that it produces half of what gets reported as a server fault. This page explains why, and what to do instead.
The reason is a property computing hardware does not have: a network is a chain. A machine, a wall socket, a cable in a wall, a patch panel, a switch, a router, a line. Seven segments, and the result is worth what the worst of the seven is worth.
That is what makes purchases in this family so often useless. A switch costing three times more, sitting behind a cable crushed under a door, changes nothing, and the budget went into the segment that was not the problem.
This article gives the family’s ordering rule, the one-page drawing that has to precede any purchase, and the two figures that genuinely decide — which we do not put here, but whose exact location we state.
Seven segments, and you control five
Name the whole chain once and half this family’s misunderstandings disappear. The machine and its network card. The patch lead to the wall socket. The cable in the wall. The patch panel and its lead. The switch. The router. And the operator’s line.
The first five are yours: you buy them, you install them, you can replace them on a Tuesday morning. The router is often mixed — sometimes supplied, sometimes yours. The line is not yours at all and it is the only segment you have no hold over.
That split decides the strategy. On what you control, the right approach is to remove the weak points once and for all. On what you do not, the right approach is a fallback, which the pillar article covers by showing that the spare capacity lives in a phone.
It also decides responsibility, and that is the commonest conversation during a fault. A network supplier cannot answer for the line; an operator cannot answer for your cabling. Between them the boundary is physical: it is the socket where the line arrives.
Write that boundary down somewhere, with a photograph. On the day of a fault, the first half hour usually goes on deciding who should be called, and that half hour is entirely avoidable.
The chain is worth its worst segment
It is the property separating this family from every other in the pillar, and it has a direct consequence for the order of purchases: there is no point improving a segment while an earlier one is worse.
The typical case we meet: a recent switch, a fast line, new machines, and cabling laid twelve years ago by an electrician who used what was to hand. The measured result is the cabling’s, and none of the other three purchases changed it.
The ordering rule follows and it is the reverse of the usual commercial order: check the cheapest segment first, the cable, then work up. A seller starts with the switch because it is an object with a price; a diagnosis starts with the cable because that is where the chain breaks.
There is an exception worth knowing: when the number of machines exceeds what the switch can carry, or when equipment needs power over the cable, the weak link genuinely is the active equipment. The page on switches handles that case and it is recognised by counting, not by guessing.
The wording we use with a client is simple: tell us what you think you are buying, and we will tell you which segment will make it useless. In more than half the cases, the answer is a cable costing a few hundred dinars.
The one-page drawing, before any purchase
Before buying anything in this family, a drawing is needed. Not an architect’s plan: a sheet showing what is connected to what, with approximate lengths and the number of sockets per room.
That drawing costs an afternoon and it changes three things. It reveals the segments nobody counted — the socket behind the cupboard, the extension crossing a corridor, the small switch put under a desk as a stopgap two years ago and never removed.
It then allows honest sizing: how many ports are genuinely occupied, how many sockets are dead, and how many would be needed to remove the extensions. A quotation drawn up without that count is a quotation drawn up from the supplier’s catalogue.
And it serves on the day of a fault, which is its main use. The question "what is plugged into this port" has no answer without a drawing, and hunting for it by unplugging cables to see what falls over is exactly the method this document removes.
Keep it current the only way that works: a photograph of the panel and the cabinet after every visit, dated, in the same folder. A schematic redrawn neatly once a year is worth less than three photographs taken at the right moment.
Label both ends
It is the cheapest measure in the whole IT pillar and the one with the highest return: every cable carries an identical label at both ends, and every wall socket carries the same number.
Without labels, every diagnosis begins with a manual operation: unplug a cable and see who loses the network. That takes an hour, it interrupts other people, and it starts again at every incident because it leaves no trace.
With labels, the same question is settled in thirty seconds. The arithmetic is easy: one afternoon of labelling once, against an hour per incident, for an estate producing ten to fifteen network incidents a year.
The method is mundane and that is exactly why it gets skipped. Simple numbering — room and socket number — written on a label that survives, at both ends of the cable, plus the same number on the patch panel. Nothing else, and above all no coding system nobody will remember.
The moment it happens almost free is installation. Doing it afterwards costs a day; asking the installer for it at quotation time costs nothing, and it is the most profitable line you can add to a cabling order.
What gets reported as a server fault
A machine that no longer reaches a shared resource always produces the same sentence: "the server is down". It is almost always false, and separating the cases takes two minutes.
The deciding test is what the support article calls elimination by comparison: can anybody else do it right now? If yes, the server is fine and the problem is on one person’s path. If no, move up one segment.
The real causes, in the order we meet them: a lead unplugged or half seated, a dead switch port, a damaged wall cable, an address conflict, and only then the server itself. The first four are network and cost almost nothing.
There is one case that fools everybody and has to be named: the intermittent fault. A slightly damaged cable works, then does not, then works again. It gets blamed on the software, the server, the user — never on the cable, because a cable is perceived as binary and is not.
The consequence is a diagnostic rule to give any supplier: on an intermittent, non-reproducible fault, change the lead before looking anywhere else. It costs a few hundred dinars and removes a large share of the long investigations.
Wireless does not replace cable
The boundary has to be set clearly, because the question comes up on every project and it is asked backwards: the right question is not "can everything be wireless" but "what has to stay cabled".
Three things have to stay cabled, and the list is short. Whatever never moves: a server, a fixed till, an accounting machine. Whatever carries a lot, regularly: a machine working on heavy shared files. And whatever has to be reliable to the minute, like a payment terminal.
The rest can be wireless without qualms, and most of an office’s comfort comes from there. Wireless is an excellent convenience and a mediocre transport medium: it shares a common space, it crosses load-bearing walls badly, and it degrades when the neighbour installs theirs.
We are not reopening the professional-versus-consumer question here: the article on professional wireless handles it, and it also explains why adding boxes does not fix a coverage problem. This page carries one rule only, and it is the boundary above.
One thing is worth adding at this level: wireless depends on the cable feeding it. An access point at the end of a bad cable broadcasts a bad network, which closes this section back onto the chain rule — the wireless segment inherits everything before it.
Two figures decide, and they are on the child pages
At this point an article in this family usually prints tables: cable categories, throughputs, power levels. We put none here, and it is not a refusal to quantify.
The two numbers that genuinely decide in this family are real, published by standards bodies, and dated. The first is the maximum length of a copper link, which sets where the cabinets go before anything is bought. The second is the power available per port to feed equipment over the cable, which decides what a switch can carry.
They are not here because they argue elsewhere. The first is on the page about installation and cabling, where it governs the floor plan. The second is on the page about switches and routers, where it governs the counting of cameras and access points.
The reason for that separation is editorial and it is the marketing pillar’s: in a decision article, a table of figures invites the reading that is false — that a network is chosen on specifications. At this level what decides is a drawing and a label, and a figure well placed on a child page beats a decorative table here.
What this page asserts instead fits in a sentence, and it is checkable at your premises: in most installations we audit, the limiting segment is a cable or a socket rather than a piece of equipment, and it costs less than the quotation that was about to be signed.
Addressing, and why two networks meet badly
Two addressing problems produce a noticeable share of unexplained faults, and both are invisible to a user, who only sees that "it stopped working".
The first is a conflict: two devices carry the same address, because one was set by hand by somebody and the other received it automatically. The symptom is intermittent and it affects both devices, which makes the cause hard to see.
The rule that avoids it is administrative rather than technical: fixed addresses are reserved in the router, not typed onto the devices, and the list of which are fixed sits in the same table as the inventory. An address fixed by hand and not written down is a deferred fault.
The second problem appears when two networks meet: remote access from an employee’s home, a second branch, a link between two sites. If both networks use the same address range — and they do, because every device leaves the factory with the same one — the link does not work.
It is prevented by a five-minute decision at installation: choose a range different from the default. That costs nothing on installation day, and it costs a full reconfiguration on the day you want to link two sites, which is to say at the moment you need it.
Separating what has no business together
On a flat network, everything sees everything. The cameras, the till, the accounting machines, a visitor’s phone on the wireless and the printer share one space, which is convenient and rarely desirable.
Three separations are worth the effort and the first is the most urgent: guests. A network open to visitors has to be separated from the working network, and that is a function even modest equipment offers today. It takes ten minutes and removes a whole category of problems.
The second covers cameras and surveillance equipment. They are numerous, they transmit continuously, they are rarely updated, and they only need to talk to a recorder. Leaving them on the working network gives them access they have no use for.
The third covers taking payment where it exists. It only needs to reach a few destinations, it is the most sensitive, and its stopping costs the most. Separating it is as much a reliability measure as a security one.
What we do not recommend is carving a twenty-person business into eight segments. Every separation adds a rule to maintain, and a configuration nobody understands six months later produces more faults than it prevents. Three separations, no more, and written on the drawing.
The operator’s line: what you can require
It is the one segment you do not own, and the relationship with it turns on what you can demonstrate rather than on what you assert.
What you can require: that the problem be observed at the delivery point, which is to say at the socket where the line arrives, independently of your installation. That is why you need to know how to plug one device directly in there and nothing else — it is the test that ends the discussion.
What you cannot require: a throughput guarantee on a consumer offer, or a visit outside the terms of the contract you signed. A business offer costs more precisely because it contains a commitment, and plenty of companies pay for a consumer offer while expecting a business service.
What genuinely helps in a complaint is a record: a measurement taken at the same place, at the same hours, several days running, noted with date and time. Without a record, an intermittent outage is your word against a technical log; with one, it is a case.
Finally, the decision that is yours alone: the fallback. If three of your functions have to survive an outage, they have to have been tested on a mobile connection on a day of your choosing. The IT pillar article says so, and this page only adds that a fallback gets tested with the same rigour as a backup.
What to obtain from an installer
A network installation gets judged three years later, which is far too late to negotiate. Five requirements, set at quotation time, make that judgement unnecessary.
Labelling at both ends and on the panel, written into the quotation as a line rather than assumed. That is section 4 and it is what separates a professional installation from one that works on day one.
Link certification, meaning a measurement taken with an instrument on every socket installed, with a printed result. A serious installer offers it; one who answers that "it works" is offering a test rather than a measurement, and the difference shows on long runs.
The drawing handed over at the end of the job, with socket numbers matching the labels. Without that document the labelling itself loses half its value, because nobody knows where number 14 goes.
Finally, two physical things: spare sockets, because an extra socket installed during the job costs a fraction of its price installed two years later; and a route that stays accessible, because a cable buried in concrete does not get replaced and this family gets replaced every fifteen years.
What we do, and what we will refuse to do
What we will refuse: selling active equipment before checking the cable and sockets in front of it. It is the easiest quotation in this family and the one that leaves the problem exactly where it was.
We will refuse to hand over an installation without labels at both ends and without a drawing. An unlabelled installation works on day one and costs an hour at every incident for fifteen years, which makes it the only saving in this pillar that gets paid for every year.
We will refuse to carve a twenty-person network into eight segments because a catalogue recommends it. Three separations — guests, surveillance, payment — and written on the drawing.
What we do: the chain named segment by segment and the boundary with the operator photographed; the one-page drawing before any purchase; labelling carried into the quotation as a line; the lead replaced before any investigation of an intermittent fault; the address range chosen away from the default at installation; and links measured rather than tried.
And what you can do this week without us: open your network cabinet, photograph it, and count how many cables carry a readable label at both ends. That number, against the total, predicts the length of your next network fault fairly well.
Frequently asked questions
Should we change the switch to go faster?
Rarely. A network is a chain and it is worth what its worst segment is worth: in most installations we audit, the limiting segment is a cable or a socket. Check the cheapest before buying the dearest.
Why label cables?
Because without labels every diagnosis starts by unplugging cables to see who loses the network — an hour, at every incident, for fifteen years. Label both ends and the panel, and ask for it in the installation quotation.
Can everything go wireless?
The right question is what has to stay cabled: whatever never moves, whatever carries a lot regularly, and whatever has to be reliable to the minute. The rest can be wireless. And an access point at the end of a bad cable broadcasts a bad network.
Why does this page give no technical tables?
Because the two deciding figures argue elsewhere: the maximum copper link length on the cabling page, where it governs the floor plan, and power per port on the switches page, where it governs the counting of cameras.
Should the network be separated?
Three separations are worth it: guests, surveillance, and payment. No more in a twenty-person business — every separation adds a rule to maintain, and a configuration nobody understands produces more faults than it prevents.
How do we handle an intermittent operator outage?
By building a record: a measurement at the delivery point, at the same hours, several days running, noted with date and time. And by testing a mobile fallback on a day you choose rather than one chosen for you.
Where we come in
The number of legible labels in your cabinet predicts most of the rest. It does not say which segment is actually slowing you down.
- We follow the chain from the desk to the operator’s socket, photographing it.
- We test the passive side before proposing you any new box.
- We leave a diagram another firm could read without phoning us.
Fewer than ten desks and nobody complaining of slowness: change nothing, we will find nothing useful to sell you.
Read next
The line is down: establishing which side, and what comes next
Five minutes decide whether the fault is yours or the operator’s. What to record before calling, how to escalate, and what a second line really covers.Moving the business: what to order three months ahead, and the weeks you exist at two addresses
A move tests your whole system at once, on a date a landlord set. The long pole is never the van.Telephony: the number is the asset, not the system
Your customers know a number, not a piece of equipment. Everything decided here has to protect that number and what happens when somebody dials it.
Let us talk about your project
A free audit, no commitment: we look at your online presence and tell you what is holding it back.