Route Analysis

Georgia to Taiwan: 743ms Through Nine Countries — Why Cogent Sends Packets Across North America to Reach Asia's Biggest Cable Hub

Published: March 16, 2026

Taiwan is one of the most connected places on Earth. With over 20 submarine cables landing on its shores — including FASTER, APCN-2, Asia Pacific Gateway, and the New Cross Pacific Cable — it sits at the crossroads of virtually every major trans-Pacific and intra-Asian network. Any data packet in the world should be able to reach Taipei in just a few hops.

From our probe in Tbilisi, Georgia, it takes 30 hops, crosses nine countries, traverses all of North America from east to west, and arrives — on a bad day — in nearly two full seconds. Over 551 measurements, the average round-trip time is 322ms, but individual measurements have hit 1,922ms. And the packet's journey includes one of the strangest routing anomalies we have ever recorded: after arriving in Taiwan, the data bounces back to Los Angeles before returning to Taipei.

The Route: Tbilisi to Taipei via Montreal, Chicago, and San Jose

Here is the hop-by-hop breakdown of a typical measurement (743ms) from our Tbilisi probe to Taipei:

Hop	Location	Operator	RTT
1–7	Tbilisi, Georgia	JSC Global Erty	25ms
9	Belgrade, Serbia	Cogent Communications	35ms
10	Vienna, Austria	Cogent Communications	53ms
11	Munich, Germany	Cogent Communications	50ms
12	Frankfurt, Germany	Cogent Communications	60ms
13	Paris, France	Cogent Communications	159ms
14	Frankfurt, Germany	Cogent Communications	69ms
15	Paris, France	Cogent Communications	160ms
16	Montreal, Canada	Cogent Communications	101ms
17	Toronto, Canada	Cogent Communications	165ms
18	Montreal, Canada	Cogent Communications	158ms
19	Toronto, Canada	Cogent Communications	164ms
20	Carter Lake, USA	Cogent Communications	174ms
21	Chicago, USA	Cogent Communications	172ms
23	San Francisco, USA	Cogent Communications	193ms
24	San Jose, USA	Cogent Communications	200ms
25	Palo Alto, USA	Cogent Communications	209ms
26	Palo Alto, USA	HiNet Service Center in U.S.A	207ms
28	Taoyuan City, Taiwan	—	339ms
29	Los Angeles, USA	HiNet Service Center in U.S.A	286ms
30	Taipei, Taiwan	—	743ms

The route reads like a geography lesson gone wrong. From Georgia, the packet crosses the Balkans to Serbia, zigzags through Central Europe (Vienna → Munich → Frankfurt → Paris → back to Frankfurt → Paris again), jumps the Atlantic to Montreal, bounces between Montreal and Toronto, then works its way across the entire United States: Chicago → Carter Lake → San Francisco → San Jose → Palo Alto.

But the most remarkable moment comes at the end. At hop 28, the packet arrives at Taoyuan City, Taiwan — the location of Taiwan Taoyuan International Airport and a major Chunghwa Telecom facility. Then at hop 29, it appears in Los Angeles. Then at hop 30, it returns to Taipei at 743ms.

The Boomerang: Why Does the Packet Return to the US?

The Los Angeles hop after Taiwan is not a measurement error. HiNet — Taiwan's dominant telecommunications provider, operated by state-owned Chunghwa Telecom — maintains a subsidiary called "HiNet Service Center in U.S.A" (AS9680) based in San Jose, California. This US presence handles HiNet's international peering and transit.

What appears to happen is this: the packet arrives at a Chunghwa Telecom facility in Taoyuan via the transpacific cable. But the ICMP response (the traceroute reply) is generated by a management interface that routes back through HiNet's US operations center in Los Angeles. The actual data traffic likely stays in Taiwan — but the control plane bounces back across the Pacific, adding what appears to be a round trip but is actually an artifact of HiNet's network architecture.

This pattern appeared in our worst-case measurement as well (1,922ms), where the same Taoyuan → Los Angeles → Taipei bounce is visible. The final hop in Taipei showed 1,922ms — a measurement that likely reflects severe congestion at HiNet's internal routing, not an actual journey time.

The Paris-Frankfurt Oscillation

Something else unusual happens in the middle of the route. Look at hops 12–15:

Hop	Location	RTT
12	Frankfurt, Germany	60ms
13	Paris, France	159ms
14	Frankfurt, Germany	69ms
15	Paris, France	160ms

The packet bounces between Frankfurt and Paris twice. Frankfurt shows consistent low RTT (60–69ms), while Paris shows dramatically higher RTT (159–160ms). This is almost certainly ICMP rate limiting at the Paris router — the actual forwarding path goes Frankfurt → Paris → transatlantic, but the Paris router is heavily loaded and responds slowly to traceroute probes. The real transit delay from Frankfurt to Paris is about 8ms, not the 100ms the numbers suggest.

A similar oscillation appears between Montreal and Toronto (hops 16–19), where the packet bounces back and forth before finally heading west. This reflects Cogent's internal routing decisions — possibly load balancing between their Montreal and Toronto points of presence before selecting the optimal westward path.

Three Probes, Three Completely Different Paths

The most striking finding is how dramatically the route changes depending on the source:

From Tbilisi (Georgia): 9–10 countries, via Cogent across North America. Average 352ms.
Path: GE→RS→AT→DE→FR→GB→CA→US→TW

From Almaty (Kazakhstan): 6 countries, via Russia and Germany. Average ~335ms.
Path: KZ→RU→DE→TW

From Minsk (Belarus): Two distinct routes observed:

Via Hong Kong: BY→RU→DE→HK→TW (213 measurements)
Via Australia: BY→RU→DE→AU→TW (100 measurements)

The Almaty route is the shortest and most logical — it follows the Eurasian landmass east through Russia, then through the Trans-Europe-Asia cable infrastructure to Frankfurt, and from there directly to Taiwan. The Minsk routes are interesting because they split: most traffic goes through Hong Kong (likely via the FLAG or SEA-ME-WE cables), while a significant portion inexplicably routes through Australia.

But the Tbilisi route is the paradox. Instead of following the Almaty pattern through Russia and Central Asia, Cogent's network in the Caucasus feeds traffic westward into Europe and then across the Atlantic — adding an entire ocean crossing that other probes avoid entirely.

20+ Cables, One Bottleneck

Taiwan's cable infrastructure is extraordinary. Our database shows over 20 international submarine cables landing at four different locations:

Toucheng alone hosts APCN-2, Asia Pacific Gateway, Apricot, Candle, E2A, ORCA, NCP, PLCN, FLAG North Asia Loop, and SJC2. Tanshui has APCN-2, EAC-C2C, FASTER, and SJC2. Fangshan has EAC-C2C and SJC2. These cables connect Taiwan directly to Japan, Hong Kong, Singapore, the Philippines, Australia, and the United States.

Yet none of this infrastructure helps traffic from Georgia. The bottleneck is not Taiwan's connectivity — it is Cogent's routing policy. Cogent (AS174) carries the packet from Tbilisi to San Jose using only its own network. It does not peer with Asian transit providers in Europe. It does not use the Trans-Siberian terrestrial route that Almaty's traffic follows. It sends everything westward across the Atlantic to North America, then eastward across the Pacific.

The handoff to HiNet happens in Palo Alto, California — not in Frankfurt, not in London, not in Hong Kong. This is where Cogent and HiNet peer. And this single peering point determines the entire route geometry.

The Business Logic

Why does Cogent route through North America when a direct path through Asia would be shorter? The answer lies in Cogent's network architecture and business model.

Cogent is one of the world's largest Tier 1 internet providers, but its backbone is fundamentally North America-centric. The company's core network spans 56,000 miles of fiber across North America and Europe, with major hubs in Washington DC, New York, Chicago, Dallas, and Los Angeles. Their European presence extends to Frankfurt, London, Paris, Amsterdam, and Milan.

Cogent does not operate its own infrastructure across Asia. To reach Taiwan, it must hand traffic to another carrier — and it chooses to do this in the United States, where it has the densest peering arrangements. The HiNet peering point in Palo Alto gives Cogent a clean handoff to Taiwan's national carrier, and HiNet takes responsibility for the transpacific leg via its own cables (FASTER, or the cross-Pacific capacity it purchases from other operators).

For Cogent's customers in the Caucasus, this means every packet to East Asia makes a full transatlantic crossing before reaching a peering point. But for Cogent's business model — which emphasizes using its own infrastructure to minimize transit costs — the route makes financial sense even if it doubles the latency.

What the Data Tells Us

Our 551 measurements reveal several patterns:

The most common path from Tbilisi is GE→RS→AT→DE→FR→GB→CA→US→TW (35 measurements, avg 352ms), but the route is far from stable. We observed at least 10 distinct country-level paths, all going through North America but varying in which European countries the packet traverses.

The RTT range is enormous — from 8ms (likely a cached or local response) to 1,922ms. The median is around 350ms, but measurements above 500ms are common, suggesting that congestion at the Cogent-HiNet peering point or within HiNet's network is a recurring issue.

The contrast with Kazakhstan is the most telling data point. From Almaty, the path KZ→RU→DE→TW is direct and consistent, achieving roughly similar latency with a fraction of the geographic distance. The difference is the upstream provider: Kazakhstan's transit goes through Transtelecom and RETN, which have direct European-Asian interconnections that Cogent lacks.

For a destination with 20+ submarine cables and connectivity to virtually every global network, reaching Taiwan from the Caucasus should not require crossing nine countries and two oceans. The infrastructure exists. The cables are in the water. The problem is that the routing economics of Tier 1 providers do not always align with geographic efficiency — and for traffic from the Caucasus, Cogent's path through North America remains the only path available.

Data based on 551 traceroute measurements from GeoCables probes in Tbilisi (Georgia), Almaty (Kazakhstan), and Minsk (Belarus) collected between February 24 and March 16, 2026, using RIPE Atlas infrastructure.