Linking two distant company sites in Algeria (2027): VPN between head office, branches and warehouse, address plan, speed and backup, for when each site works on its own

“We have a head office in Algiers, a branch in Oran and a warehouse in Sétif. The business software is at head office; the branch sends its sales by file in the evening, or works through TeamViewer on a PC we leave switched on. Prices are not the same everywhere, the warehouse stock is known a day late, and when the branch's connection drops it can do nothing.” Linking sites is not buying “a VPN”: it is an address plan that does not overlap, a firewall at each end, an encrypted tunnel that comes back by itself, and the right choice of what travels: the screen, a few requests or a copy.

Symloop13 min read
Linking distant company sites in Algeria: tunnels, addresses, backupAddress plans with no overlap, a firewall at each site, encrypted tunnels, software that gets throughLINKING COMPANY SITES · ALGERIA · 2027Linking distant company sitesin Algeria: tunnels, addresses, backupAddress plans with no overlap, a firewall at each site,encrypted tunnels, software that gets throughEACH SITE ON ITS OWNfiles sent in the evening, different pricesONE NETWORK, THREE SITESa tunnel that comes back by itselfsymloop.com

Short answer. Linking distant sites in Algeria in 2027 holds five things. An address plan: one block per site, with no overlap, because two sites carrying the same addresses cannot be linked. A firewall at each site, with at least one fixed point, head office or a hosted central point, towards which the others open their tunnel. Encrypted tunnels across the internet, one per branch, re-established by themselves. Rules between sites: the branch reaches the server and telephony, not the whole head office. And the way each piece of software crosses, remote desktop, web application or synchronised local copy, chosen from the measured delay, with a backup link. This guide gives the drawing, the calculation that explains why software becomes unusable at a distance, and the six decisions to make before linking a site.

It is for the manager who has opened a second address, the company whose branches, points of sale or warehouse each work on their own, the one that depends on a remote-control tool to keep its branches running, and the founder living in France who wants to see their Algerian company as if they were there. The marks quoted are those of public references: private address ranges, IPsec and WireGuard tunnels, the default port of remote desktop. The sites, speeds, delays and number of exchanges in the example are constructed to show the reasoning, not surveyed at a client's. Operator offers, fixed public address, private link between sites, guaranteed speeds, depend on your addresses and are checked with them; their prices and contracts are outside the scope of this guide.

Three sites, two tunnels: head office, branch and warehouse on one network

The figure draws the Algiers head office, the Oran branch and the Sétif warehouse, and what passes between them.

Figure 1 · The Algiers head office, the Oran branch and the Sétif warehouse linked by two encrypted tunnels across the internet. Each site has its own address plan; the head office has a fixed public address and receives the tunnels the other two open. Throughput between two sites is capped by the smaller of the two, upload of the sender and download of the receiver: Sétif sends at only ten megabits per second. Before, the three sites carried the same factory address plan and could not be linked. When the Oran link drops, the tunnel comes back over the backup in twenty-five seconds. Speeds, delays and addresses are a constructed example.

The address plan: company networks use private address ranges, the same all over the world; each site's operator box arrives with the same factory setting, so head office, branch and warehouse often carry exactly the same addresses. Linking two overlapping networks creates duplicates: a head-office workstation and a branch workstation have the same address, and no tunnel knows which to talk to. The example gives each site a block, 10.20 for Algiers, 10.21 for Oran, 10.22 for Sétif, each with the same separated networks as in the guide on network installation. The fixed point: a tunnel opens towards an address that can be reached; head office therefore has a fixed public address, and the branches are the ones opening the tunnel towards it. When no site can have one, or the branches sit behind an address shared by the operator, the fixed point is placed on a small hosted server.

The tunnels: each branch sets up an encrypted tunnel to head office across the internet; what flows through it is unreadable to anyone but the two firewalls. The two common techniques are IPsec, a protocol suite that authenticates and encrypts packets, and WireGuard, more recent, part of the Linux kernel since 2020; the choice depends on the firewalls in place. Throughput between two sites is capped by the smaller of the two, upload of the sender and download of the receiver: the Sétif warehouse receives at twenty megabits per second but sends at only ten, which matters for its backups and cameras. Delay is measured: twenty milliseconds between Oran and Algiers in the example, sixteen between Sétif and Algiers. Backup: when the Oran link drops, the firewall moves to the second link and brings the tunnel back, in twenty-five seconds in the example; telephony between sites travels through the same tunnel.

01

Why the software is slow at the branch: 400 round trips, not the bandwidth

The figure opens the same screen five ways and counts the seconds.

Figure 2 · The same screen opened five ways. Software installed on the workstation exchanges four hundred times with its database to display one record: at one millisecond per round trip, that is four tenths of a second at head office; at twenty milliseconds, it is eight seconds at the branch, whatever the bandwidth. By remote desktop, through a web application or with a synchronised local copy, the same screen comes back in under a second. Number of exchanges and delays are a constructed example; they are measured on your software.

The calculation: much business software installed on the workstation was written for a local network; to display a customer record and its orders, it asks the database hundreds of small questions, one after the other. At one millisecond per round trip, four hundred questions take four tenths of a second, and nobody notices. At twenty milliseconds, between branch and head office, the same four hundred questions take eight seconds. Doubling the link's speed changes nothing: the delay of a round trip depends on distance and route, not on the width of the pipe. This is the most frequent reason for a tunnel that “works” and a branch that cannot get its work done.

Three ways to cross. Remote desktop: the software runs on a server at head office, and only screen, keyboard and mouse travel; the screen comes back in under a second, with nothing rewritten, at the cost of a server to size and maintain. Web application: the browser sends a few requests and the server does the rest; this is the form software written today takes, and the best one when the software has to be replaced anyway. Synchronised local copy: the branch reads and writes on site, and changes go to head office in the background, as in the guide on offline work; it is the only one that carries on when the link drops, which makes it the choice for a till or a warehouse. The choice is made software by software, after measuring the delay and counting the exchanges, and it joins what an application designed for weak networks does.

«Screen time is the number of exchanges multiplied by the delay. A faster link does not change the delay; four hundred round trips at twenty milliseconds make eight seconds.»
02

Six real situations, and what linking the sites must do

Each row is a sentence heard in a multi-site company, and the answer is a network or software decision.

Figure 3 · The first three rows are settled by a tunnel and by the right way for the software to cross; the next three by closing the open ports, a second link and one address plan per site. None is settled by a faster internet subscription.

The branch sends its sales by file in the evening: head office decides in the morning on yesterday's figures; with one piece of software reached through the tunnel, the Oran sale is visible in Algiers within the minute, and the price is the same everywhere because it is written in one place only. The software is unusable at the branch: that is the calculation in figure 2; the way of crossing is changed, not the link. We work through TeamViewer on a head-office PC: a remote-control tool is made for fixing a problem, not for running a branch; it ties up one head-office PC per person and depends on an outside service and a password that goes round. Through the tunnel, each person has their own account and session.

Remote desktop is open on the internet: a port forward set up one day to fix something leaves the remote desktop service, which listens by default on port 3389, reachable by anyone; servers exposed this way have been compromised and then resold, according to researchers quoted by public references. With a tunnel, no port is open any more: the only way in is the tunnel, with one's account and two-step sign-in. The branch link drops and it can do nothing: a second link and a tunnel that comes back by itself cover the short outage; a till that carries on offline covers the long one. The sites cannot be linked: all three carry the same address plan; one plan per site is written, renumbering is done one evening, then the tunnels are set up.

03

Linking sites, in five layers

Encrypted tunnels are the accented layer: the address plan and the firewalls make them possible, the rules and the way of crossing decide what they carry.

Figure 4 · The accented layer is the encrypted tunnels: it is what turns three networks into one without renting a private line. The address plan and the firewalls, above, make it possible; the rules between sites and the way of crossing, below, decide what it carries and what it refuses.

Address plan. One block per site, written once, with room for sites to come; the same division everywhere, workstations, telephony, servers, cameras, guests, so that a rule written for one branch holds for all. Firewall per site. The same model or family everywhere simplifies tunnels and monitoring; a four-workstation branch does not need head office's firewall, but it does need a real firewall, not the operator box. Encrypted tunnels. In a star, everything goes through head office: two branches need two tunnels, ten branches ten; it is simple, and head office becomes the point everything depends on, hence its second link and its UPS. In a full mesh, each site talks to every other: three tunnels for three sites, fifteen for six, forty-five for ten; only what really exchanges a lot gets a direct link.

Rules between sites. A tunnel is not an open door: the branch reaches the software's server, the shared printer and telephony, not head office's cameras nor the accounting workstations; an infected workstation at the branch must not be able to wander around head office. Access for individuals, a director on the move or an accountant at home, goes through personal remote access with two-step sign-in, as in the guide on roles. Crossing and backup. Each piece of software has its way of crossing written down; each site has a second link when its activity cannot stand a stop; delay, speed and tunnel state are monitored, and an alert goes out when a tunnel drops, on the logic of the guide on monitoring.

«A tunnel is not an open door onto the whole head office: the branch reaches the server and telephony, and nothing else.»
04

Linking sites in 2027, in weeks

Three weeks in the example for three sites, the first with nothing changed.

Figure 5 · Three weeks in the example for three sites, the first with nothing changed: the survey of addresses, the link measurements and the plan are decisions. The accented week is addresses and tunnels, because renumbering a site is prepared and done in one evening. A private link rented from an operator, where it exists at your addresses, follows its own lead time.

Week 1, survey and plan: addresses of each site, equipment, links and their real speeds, delay measured between sites at several times of day, list of the software that must cross and number of exchanges for a typical screen, address plan, shape of the network, request for a fixed public address from the operator if needed. Week 2, addresses and tunnels: overlapping sites are renumbered, one evening each; firewalls are installed; tunnels are set up and tried with two pilot workstations while the rest works as before.

Week 3, software and backup: each piece of software is switched to its mode, remote desktop, web application or local copy; the port forwards and remote-control tools that served as the link are removed; the outage of a branch link is provoked, and the time for the tunnel to come back is noted. Two nearby sites are linked in a few days; ten points of sale, sites without a fixed address or software to rewrite as a web application take longer. A private link rented from an operator, where it exists at your addresses, has its own lead time and is decided with them.

05

Six decisions before linking your sites

The first is accented: without a written address plan, each new site brings its factory setting.

  • What the manager sees: the state of each tunnel, delay and speed per site, sales and stock of every site in the same software, an alert when a site is cut off.
  • What the company no longer does: send files in the evening, leave a PC switched on for each branch, open a port on the internet, discover that a price differs from one site to another.
  • What is measured: delay and speed between sites, time to open a typical screen at the branch, time for the tunnel to come back over the backup, ports open on the internet, which must stay at zero.
Figure 6 · The accented decision is the first: without a written address plan, each new site brings its factory setting and blocks everything. The other five then follow in order: the shape of the network, the fixed point the tunnels open towards, the rules between sites, the way each piece of software crosses, and the backup.

Write the address plan: one block per existing and future site, recorded in the network file. Star or mesh: the star as long as everything goes to head office; a direct link between two sites only if they exchange a lot with each other. Find a fixed point: a fixed public address at head office if the operator provides one; otherwise a name that follows the address when it changes; otherwise a hosted central point, which has the advantage of depending on no site's link.

Write the rules: for each site, what it reaches at head office and what it does not, and who has personal remote access. Choose the crossing: for each piece of software, the screen, a few requests or a copy; this is where it is decided whether a remote desktop server is needed, a web application to be written, or a synchronisation. Plan the backup: which sites have a second link, what carries on offline, who receives the alert. These six decisions fit on two pages and hold for the tenth site as for the second.

How do you link two distant company sites?
In four steps: give each site an address plan that does not overlap the other; install a firewall at each site; set up between them an encrypted tunnel across the internet, opened by the branch towards head office; write down what each site is allowed to reach. Only then do you choose how each piece of software crosses, and add a backup link where a stop is costly.
VPN or the operator's private link: which one?
The encrypted tunnel uses your existing internet links: it is set up quickly and follows your changes of operator; its quality is that of your links. A private link rented from an operator offers contractual speeds and delays, where it is available at your addresses, with its lead time and contract. Many companies start with tunnels, measure, then ask for a private link for the one site that needs it.
Do we need a fixed IP address?
You need a point the other sites can reach. A fixed public address at head office is the simplest, depending on what the operator offers. Failing that, a dynamic name follows the address when it changes. If the sites sit behind an address shared by the operator, as is often the case on 4G, they cannot receive an incoming connection: the meeting point is then placed on a small hosted server, towards which every site opens its tunnel.
Why is our software slow through the VPN?
Almost always because of the number of exchanges, not the bandwidth. Software written for a local network asks its database hundreds of questions for a single screen; each one waits for the round trip, twenty times longer between two cities than inside a building. It is checked by counting the exchanges of one screen and measuring the delay. The remedies are remote desktop, a web application, or a synchronised local copy.
Can we carry on with TeamViewer or AnyDesk?
For fixing a workstation, yes. For making a branch work every day, no: it needs one head-office PC switched on per person, the session depends on an outside service, and access gets shared. A tunnel between sites with one account per person does the same without those dependencies, and lets you know who connected.
Is it dangerous to open remote desktop to the internet?
Yes. A remote desktop service reachable from the internet is tried continuously by bots, and servers exposed this way have been compromised and then resold, according to researchers quoted by public references. The rule is to open no port: you first come in through the tunnel or through personal remote access with two-step sign-in, and remote desktop is reachable from the inside only.
How many sites can be linked?
As many as you like in a star: each new site adds one tunnel to head office. In a full mesh, the number of tunnels grows fast: three for three sites, fifteen for six, forty-five for ten. So the star is kept, with a well backed-up head office or central point, and a direct link is added only between two sites that exchange a lot.
What happens when a site's link drops?
If the site has a second link, the firewall fails over and brings the tunnel back without intervention; twenty-five seconds in the example. If it has none, the site is isolated until the link returns: what runs on site carries on, what depends on head office stops, except software with a local copy, which resynchronises afterwards. This case is provoked before go-live so that nobody discovers it on a sales day.
Send us your sites' addresses, your operators' names and the software your branches use: we come back with an address plan and the way each piece of software will cross

Head office, branches, points of sale, warehouse, or a founder in France running a company in Algeria: we write the address plan, install the firewalls, set up the tunnels and their backup, and get each piece of software across, up to rewriting it as a web application when that is the right choice

A survey of the sites, measured delays, a renumbering one evening, an outage provoked before go-live. We work from Algiers and Sétif, in French, Arabic and English.