Hokkaido-Sakhalin Cable System (HSCS)

The Hokkaido-Sakhalin Cable System (HSCS) is a short, unshowy piece of internet infrastructure. Just 570 km of fibre between Ishikari on the northwest coast of Hokkaido, Japan, and Nevelsk on the southwestern coast of Sakhalin, Russia. The cable has been in service since 2008. It appears on very few industry maps, is almost never mentioned in press releases, and is not part of any hyperscaler consortium. Yet one direction of this ordinary 570 km cable produces the tightest, most stable round-trip measurement in our entire database.

Over the last 30 days, HSCS gave us 36 clean samples from Nevelsk toward Ishikari. The minimum was 20.3 ms. The average was 21.2 ms. The maximum was 22.1 ms. The standard deviation was 0.4 milliseconds. No other cable we monitor has a one-sided distribution that tight.

A glass line across La Pérouse Strait

Geographically, HSCS is a north-south cable. Nevelsk sits on the west coast of Sakhalin at 46.7° N. Ishikari sits on the west coast of Hokkaido at 43.2° N. Between them the fibre crosses La Pérouse Strait — the narrow body of water separating Sakhalin from Japan — and keeps going. The cable then parallels the eastern edge of the Sea of Japan before coming ashore at Ishikari, about twenty kilometres north of Sapporo. There is no branching unit, no mid-ocean junction. It is a single point-to-point fibre.

Both endpoints are on seismically active coastlines. The Sakhalin coast sees frequent quakes along the Sakhalin–Hokkaido fault system. The cable route, as a result, runs through marine protection zones where fishing trawlers are kept clear. This is common for cables in strait regions: the single biggest threat to an undersea cable is not the ocean — it is an anchor or a fishing net.

The quietest line in our data

What we measure on HSCS, from Nevelsk toward Ishikari, is the closest thing to a straight line a round-trip ping can produce. Here is a daily summary from the last three weeks:

A 0.4 ms standard deviation over three weeks means every single measurement is within about half a millisecond of every other. Pings across the internet routinely wobble by 10–50 ms at peak hours. This one does not. If you graphed the Nevelsk → Ishikari samples on a timeline, the line would look drawn with a ruler.

For comparison: the Tonga Cable, which we recently documented as the tightest ratio-to-physics in our set at 1.26×, has a standard deviation of 7.9 ms over comparable sample counts. HSCS Nevelsk → Ishikari is almost twenty times quieter than that.

Physics floor and the reality of backhaul

Light in a submarine fibre travels at roughly 204,500 km/s. For a 570 km one-way, that is 2.79 ms. Round-trip: 5.57 ms. If both ends of our ping were sitting at the landing stations themselves, 5.57 ms is the number we would expect.

We get 20.3 ms. That gap — about 15 ms of extra delay over what glass physics requires — is backhaul. Our probes are not inside the cable station racks; they sit in telecom facilities further inland, at the Internet handoff points of the local operators. From Nevelsk the signal enters the submarine cable, travels to Ishikari, then has to cover several hundred kilometres over terrestrial fibre to reach the nearest Japanese internet exchange, and the same on the way back. That "landside" portion is well-engineered, consistent, and predictable — hence the 0.4 ms standard deviation.

Why the other direction is different

Now look at the reverse. From Ishikari toward Nevelsk, the same cable, the same physical ocean crossing: 54.5 ms minimum, 61.8 ms average, 94.4 ms maximum, 7.5 ms standard deviation. Three facts about the internet become visible in that one table row:

• Routes are independently chosen per direction. The return trip from the Japanese probe to the Russian target can use whatever path the Japanese transit provider prefers, which may not be HSCS at all — it may be a longer route via Tokyo and an intercontinental hub.
• The Russian backhaul is longer. Nevelsk on Sakhalin is far from the main Russian internet backbone, which runs through Moscow via the trans-Siberian fibre. Even if the cable itself is fine, the local route from the Nevelsk landing to a routable IP may traverse many hundreds of kilometres of island and mainland terrestrial fibre.
• Asymmetric peering adds jitter. Our ping pairs are a probe at one end and a carrier-operated IP at the other. The carrier side may aggregate that address across multiple points of presence, introducing variation depending on which POP is currently serving the ping.

None of this is a fault in HSCS. The cable works. What we are seeing is the classic shape of a well-run submarine link embedded in a lopsided terrestrial environment: one side short, clean, direct; the other side longer, more contested, more variable.

A 570 km success story

HSCS is 17 years old. It predates every hyperscaler cable, every 400 Gbps coherent transponder, every modern piece of the cross-continental internet we track on this site. It has never been on the cover of a trade magazine. And yet, day after day, it delivers a 20 ms ping from Russia's easternmost populated coastline to the northern edge of Japan with accuracy better than half a millisecond.

Short point-to-point cables like this one tend to be the quietest in any monitoring dataset. They have fewer amplifiers than long-haul systems, fewer fibres to multiplex, fewer branching units, and — above all — fewer kilometres for entropy to act on. When a 570 km cable is built well, tended carefully, and carries predictable traffic volumes, the result is a measurement so consistent it stops looking like a network and starts looking like a laboratory reference.

The lesson is not that HSCS is technologically unusual. The lesson is the opposite: a well-engineered submarine cable running in its normal steady state behaves like a physics experiment. Everything above the physics floor is somebody's engineering choice. When the engineering is done right, the floor is exactly where you stop.

What we do not see

Our monitor only measures where we have a probe. From Nevelsk toward Ishikari the setup is almost ideal — short landside leg, direct cable traversal, clean Japanese internet exchange on the other side. From Ishikari toward Nevelsk the Russian side of the measurement is less controlled, and the reverse path may not even touch HSCS at all. So the 54 to 94 ms number should be read as "what our probe sees when it pings a Russian address from a Japanese probe" — not as "what HSCS delivers westbound".

We have no way, from outside, to prove which of the two return paths — via HSCS or via a longer continental route — is actually in use on any given sample. That is one of the inherent limits of ping-based monitoring across cables we do not operate. What we can say with certainty is that the cable, in the direction where we measure it cleanly, is healthy, stable, and functioning exactly the way 570 km of fibre should function in 2026.

Try it yourself

See the live data we collect on this cable on the HSCS page. For context on how short regional cables compare to long hyperscaler trunks, see our pieces on Tonga Cable, COBRAcable, and JUPITER.