Tunisia in 70 ms: six cables and the Italian Sparkle gateway

Open a browser in Tunis and ask for anything that lives outside of Tunisia. A Paris news site. A Google search. A Microsoft update. The packet leaves your machine, it runs across a fibre that reaches the roof of your apartment, it gets handed from one operator to another, and a few dozen milliseconds later an answer comes back. The end-to-end latency you feel is the sum of physics and politics: how many kilometres the glass has to carry your bit, and which companies have agreed to pass the bit along the way.

This article is the politics. We set up an experiment on 12 April 2026 — four of our RIPE Atlas probes spread across Belarus, Kazakhstan, Georgia and Israel, four real Tunisian target IPs sitting on four different operators, thirty-two measurements between them, all within a single hour. The result was clean and surprisingly consistent. The median round-trip time to Tunisia from our European and Middle-Eastern vantage points is about 70 milliseconds. And almost every single path — no matter which probe, no matter which operator — eventually funnels into the same Italian carrier. This is the story of why.

The six cables of Tunisia

Tunisia is a modest country by internet-infrastructure metrics. It has roughly 12 million people, two submarine-cable landing zones, and six cables. That is fewer cables than Pakistan (nine, all landing in Karachi), fewer than Egypt (dozens, concentrated in Alexandria and Suez), and more than any landlocked neighbour. For its size and for its role as a Mediterranean bridge between Europe and Africa, six is about right.

The two landing zones sit 130 km apart on Tunisia's north coast. Bizerte, on the north-west corner, handles the long-haul cables — the ones that cross whole oceans. Kelibia, on the Cap Bon peninsula pointing towards Sicily, handles the regional traffic to Italy. The split is almost geometric: long-haul west, regional east. Here is the inventory our database shows, oldest cable first:

Four things stand out in that table. First, every one of the three short Italian-facing cables — KELTRA-2, Hannibal, Didon — is effectively owned by a Tunisian operator plus, in the case of KELTRA-2, the Italian carrier Sparkle. Second, Sparkle also sits on SeaMeWe-4 as one of the consortium members. Third, the newest cable on the list, Medusa, has only one owner mentioned — AFR-IX Telecom — and that one owner is not Italian. And fourth, the three Tunisian operators (Tunisie Telecom, Ooredoo Tunisie, Orange Tunisie) are split across different cables, which means the country has at least three independent international exits at the physical layer.

Remember those four observations. They will come back as an explanation once we run the probes.

Four probes, one day, thirty-two measurements

On 12 April 2026 we fired ping and traceroute from four of our own RIPE Atlas probes to four Tunisian IPs. The probes and the targets were chosen to be both diverse and verifiable:

The targets were four different Tunisian ASes so we would not accidentally look at a single-operator view:

• 193.95.66.10 — Tunisia Backbone AS2609, the Tunisie Telecom international prefix
• 41.228.62.63 — ns1.ati.tn, the authoritative name server for the .tn country-code top-level domain, operated by ATI (the Tunisian Internet Agency)
• 41.224.33.232 — www.orange.tn, Orange Tunisia's edge web server
• 41.231.5.200 — www.topnet.tn, Topnet Tunisia's edge web server

All sixteen pings came back the same way: every target blocked ICMP. Zero packets received out of three sent, every time. This is typical for edge web servers and ccTLD name servers — both classes of host tend to drop unsolicited ICMP. So we fall back on the traceroute signal: the round-trip time of the last responsive hop inside the Tunisian AS. That hop is almost always within one or two hops of the final target, inside the same backbone, so it is a good proxy for end-to-end RTT and it is what we use everywhere below.

Results: Tunisia in 70 milliseconds

Here is the summary table. All values are last-reachable-hop RTTs from the traceroute, in milliseconds. Cells are coloured nothing — read them as raw numbers.

Three of the four probes cluster tightly between 62 and 84 milliseconds regardless of target. Almaty, the only probe more than 5 000 km from Tunis, sits apart at ~125 ms. And inside each probe, the four operator targets agree on their latency to within a couple of milliseconds, which is strong evidence that the limiting factor is the path into the country, not any one Tunisian operator's own network.

Compared to physics the numbers look like this:

Now the oddity jumps out. Jerusalem is the probe physically closest to Tunis, and it is the probe with the worst multiplier. The packet travels the shortest distance yet spends the largest fraction of its time in overhead. Tbilisi, 870 km further away, gets a multiplier of 2.34× — comfortably inside what a well-provisioned Mediterranean path looks like. Jerusalem gets 3.67×. To understand why, we have to look at the hops.

Hop by hop: the four paths

Minsk → Tunisia Backbone

The path from Belarus starts locally on Beltelecom, crosses into Russian transit at hop seven, and then hands over directly to Tunisia at hop twelve.

The big visible jump at hop eight (148.7 ms on a Rostelecom router inside Moscow) is not the real latency to Tunis — it is a Russian router answering ICMP slowly while still forwarding traffic quickly. The honest measurement is hop twelve, where the packet has already reached Tunisia Backbone and sits at 71 ms. Minsk → Tunisia is clean: Beltelecom, one Russian transit hop, handed directly to Tunisie Telecom. No Italy in sight.

Tbilisi → Tunisia Backbone

Tbilisi takes the Italian door.

That is the template of what we will see again and again for the remaining probes. Tbilisi flies across GTT's European backbone into Frankfurt at around 25 ms, hands over to Sparkle at hop nine, and reaches Tunisia over Sparkle's Mediterranean infrastructure at 65 ms — a clean 2.34× over physics.

Jerusalem → Tunisia Backbone

Jerusalem is the odd case. Israel is the probe physically closest to Tunis, yet the path is the worst relative to distance.

Look at hop five: the RTT jumps from 4 ms to 47 ms inside Cellcom's own network, before the packet has left Israel. Either Cellcom is routing Jerusalem traffic internally through Tel Aviv with a long backhaul leg, or the cross-country Israeli path to the peering point is unusually expensive. Either way, by the time the packet reaches hop eight it has already used 47 ms of the 75 ms total budget — and it is only just arriving on a Sparkle router in Milan. The Mediterranean leg from Milan to Tunis adds the remaining ~28 ms at hop nine. The 3.67× multiplier lives inside Israel and on the detour up to Milan, not on the Mediterranean crossing itself. A direct Israel–Tunisia submarine cable does not exist — there is no commercial relationship to justify one — so every Israeli packet that wants to reach Tunisia goes north into Europe first, touches Sparkle in Italy, and comes back south. Geography says 2 200 km. Topology says "via Milan".

Almaty → Tunisia Backbone

Almaty is the longest path, and the one where the Italian hand stays hidden. The route goes Kazakhstan → Russian long-haul → Orange International Carriers → Tunisia, and Sparkle does not appear at all.

This is an entirely different template. The Almaty packet goes Kazakhstan → RETN (a Pan-Eurasian carrier) → Orange OpenTransit (AS5511, the French national Tier-1) and then straight into Tunisia. No Sparkle. The reason is simple and historical: Orange's international transit business has its own direct peering with Tunisie Telecom — Orange runs Orange Tunisie after all, and it owns shares in Didon and SeaMeWe-5 that land nearby. Almaty gets Orange because its upstream RETN has a natural peering with Orange in Paris. Tbilisi and Jerusalem get Sparkle because their upstreams (GTT, Arelion) have natural peerings with Sparkle in Frankfurt and Milan. Both roads end in the same Tunisian core, but the door they come through is decided thousands of kilometres upstream.

The Sparkle gateway

Three of the four traces we just looked at touched Sparkle — AS6762, formally SEABONE-NET · Telecom Italia Sparkle S.p.A. Sparkle is the Tier-1 international carrier arm of Telecom Italia, spun off as a separate legal entity in 2003 but built on backbone assets that go back to the early 1980s. Its SeaBone IP network stretches across Italy, southern Europe, North Africa, the Middle East, the Americas, and East Asia. And its presence in our Tunisia traces is not an accident.

Go back to the cable-inventory table at the top. KELTRA-2, the 209-km regional cable that connects Kelibia in Tunisia to Trapani in Sicily, is co-owned by Sparkle and Tunisie Telecom. It has been lit since 2007 and is the newest member of a Sicily–Tunisia cable lineage that goes back to the first Italcable copper links of the 1960s. SeaMeWe-4, the big long-haul route that lands in Bizerte and reaches Marseille and Singapore, counts Sparkle as one of its 16 consortium members. SeaMeWe-5, although it does not land directly in Tunisia, connects into the same Mediterranean fabric that Sparkle operates. PEACE Cable, newer and owned by a different consortium, still uses Sparkle facilities in Marseille for onward European transit.

In other words, Sparkle is not just a transit carrier that happens to peer with Tunisia. Sparkle is the physical layer for most of Tunisia's northbound capacity. When a European network wants to reach a Tunisian IP, the cheapest and most natural route is to hand the packet to Sparkle, because Sparkle is the company that actually owns the fibre that will carry it across the Strait of Sicily. Any alternative — Orange Open Transit, Level 3, Cogent — has to buy capacity from Sparkle anyway, or detour through Marseille on SeaMeWe-4. Sparkle is the gateway.

This is also why the Jerusalem–Tunis path looks worse than physics would suggest. Israel does not peer directly with Tunisia. Israeli carriers peer with Sparkle in Frankfurt and Milan. From Tel Aviv to Frankfurt is already 3 000 km of detour. Adding that to Tunisia's leg, and Jerusalem packets spend as much total time crossing Europe as they would spend reaching the US east coast. 2 200 km of straight-line distance becomes 75 ms of measured latency because the map internet traffic uses is not the map you see on a globe.

The Orange exception

There is one target in our experiment where Sparkle never appears: www.orange.tn. From every probe, the packet reaches Orange Tunisia over a completely different path. Tbilisi and Jerusalem both finish at 213.249.105.134, which belongs to AS3356 Level 3 / Lumen. Minsk finishes at 195.22.218.217, still a Sparkle IP — but before handing off to Orange Tunisia's internal backbone. Almaty again goes Orange OpenTransit (AS5511) all the way in.

Orange Tunisia is a subsidiary of Orange Group, the French incumbent. It has its own international IP plan separate from Tunisie Telecom's, and it routes international traffic through Orange's own backbone wherever possible. Level 3 appears as the fallback when the source network's peering arrangement does not directly reach Orange in Europe. The practical effect is that Orange Tunisia acts as the second door into Tunisia at the routing layer, even though physically it shares the same Bizerte and Kelibia landings. If Sparkle had an outage tomorrow, Orange Tunisia could still serve its customers — at a slight latency penalty, because the Level 3 path adds a hop, but the country would not go dark.

The Sicily paradox

After we had the four core traces above, we added a bonus probe on purpose to see the most extreme version of the Sparkle story. The town of Trapani sits on the western tip of Sicily, 150 km across the Strait of Sicily from Cap Bon and the Kelibia landing. Between Trapani and Kelibia runs the 209-km KELTRA-2 submarine cable. The fibre is in the water, it has been in the water since 2007, and its two legal owners are Sparkle and Tunisie Telecom — exactly the two carriers that end up on both sides of every single one of the traces we have seen.

A packet from a Trapani home internet connection to 193.95.66.10 should, on paper, take about 2–3 milliseconds end-to-end. KELTRA-2 is 209 km long, round-trip fibre time is about 2 ms, the rest is last-mile and router overhead. That is what physics says. Here is what we actually measured.

Sixty-eight milliseconds. From 150 kilometres away. A multiplier of about 26× over physics — more than seven times worse than the worst of our four main probes. And the reason is staring back from the trace. The packet leaves Trapani on Tiscali (AS8612), a regional Italian ISP, goes north — not south towards Kelibia — climbs all the way up to around Milan at hop five, enters Cogent's backbone at hop seven, travels across Europe on Cogent for four more hops reaching Frankfurt or Amsterdam, then hands to Sparkle at hop eleven, and finally comes back south through Sparkle's Mediterranean infrastructure — probably still via a cable that lands near Sicily anyway — to arrive in Tunisia at hop fourteen.

The KELTRA-2 cable is physically the shortest path between this probe and its target. But KELTRA-2 is a closed bilateral capacity arrangement between Sparkle and Tunisie Telecom. Tiscali does not have capacity on it. Cogent does not have capacity on it. The IP-layer cannot see the cable, so the routing table sends the packet on the 3 000-km European detour that Cogent and Sparkle do have contracts for. The cable is there, lit, and owned by an Italian carrier, but Tiscali's packet goes north-first because the only way to use KELTRA-2 is to already be on Sparkle's network.

This is the Sicily paradox, and it is the clearest single demonstration in our dataset of why a cable map and an IP routing map are different objects. Tunisia has six cables. Sicily's closest fibre lives on one of them. And a packet from the Sicilian shore still travels via Frankfurt to get across a strait you can see from the beach.

Medusa 2026 and what changes

All of this is the Tunisia of April 2026. Later this year, the picture is supposed to change. Medusa — a new 8 760-km submarine cable led by the Spanish-Catalan operator AFR-IX Telecom — is scheduled to reach full ready-for-service in 2026. Its Mediterranean footprint is the most ambitious seen in twenty years: landings at Lisbon (Carcavelos), Barcelona, Marseille, Genoa, Bizerte, Algiers, Annaba, Port Said, and Alexandria, plus a Greek branch into Athens. Unlike SeaMeWe-4 and PEACE, Medusa is not consortium-owned in the traditional sense. AFR-IX is selling capacity directly to operators and content networks on a wholesale basis, which means — for the first time since KELTRA-2 lit in 2007 — Tunisia will have a major north-bound submarine cable whose European terminus is not under Telecom Italia's control.

What that means for our traceroutes: once Medusa is lit on the Bizerte–Marseille segment, Orange Tunisia and potentially Topnet will gain a southern entry that does not go through Sparkle's Mediterranean ring. A Jerusalem-to-Tunisia packet that currently costs 75 ms by going north to Frankfurt and then south again through Sicily could, in principle, find a direct southern path through Marseille–Bizerte on Medusa and trim 15–20 ms. We do not know yet whether that will happen — it depends on how Orange and Tunisie Telecom choose to use Medusa capacity, and on whose peering relationships come first — but the physics budget is there.

This is why we are writing this article today, before Medusa lights. The numbers in the tables above are a baseline: four probes, four targets, one hour, thirty-two measurements, median ~70 ms. In six months, or in twelve, we will run exactly the same experiment. The hops and the milliseconds will tell us which carrier actually picked up the new cable first, and by how much the country moved closer to Europe.

Conclusion

Tunisia in April 2026 is well connected, by Mediterranean-Africa standards. Six cables, two landing zones, three independent operator paths, and about 70 ms to reach any of them from most of Europe or the Middle East. The multiplier against physics sits between 2.2× and 3.7× depending on where you start from — a result comfortably inside the band of what a well-engineered trans-Mediterranean path looks like.

What is unique about Tunisia is not the latency. It is the shape of the path. Three of every four European packets that enter the country pass through the same Italian carrier — Sparkle, the Tier-1 descendant of Telecom Italia's 1970s Mediterranean cable empire. Sparkle is not a curiosity. It is the structural reason Tunisia's RTT from most of Europe looks the way it does. Understanding it is understanding why the cable map and the routing map are two different maps — one shows the glass, the other shows the politics — and why a country can have six cables and still be effectively connected through one carrier. Medusa 2026 is the first real opportunity in twenty years for those two maps to drift apart again. We will watch the second map with interest.

Try it yourself

If you have a RIPE Atlas account, you can reproduce this experiment in under five minutes of credit. The four targets we used — 193.95.66.10, 41.228.62.63, 41.224.33.232, 41.231.5.200 — will all accept traceroute even when they drop ping. Pick any three probes in Europe or the Middle East, fire a one-off traceroute with ICMP protocol, and read the last-reachable-hop RTT. If your probe's upstream peers with Sparkle, you will see the 195.22.x.x block show up somewhere between hop six and hop ten. If your probe's upstream peers with Orange OpenTransit, you will see 193.251.x.x or Level 3's 213.249.x.x instead. Either way, the last two hops inside Tunisia will sit on AS2609 (Tunisie Telecom Backbone) or on AS37492 (Orange Tunisia).

Our submarine cable map shows all six Tunisian cables, their owners, and their current operational status. The individual pages for SeaMeWe-4, Medusa, PEACE, and KELTRA-2 have their latest health measurements and RTT history. We will add a follow-up article when Medusa lights its Tunisia segment — watch this space.