Another cable break: check your own routing in 10 minutes, and know when buying more actually helps

On 25 August, the SJC2 and AAE-1 submarine cables both failed while AAG remained unrepaired — three of Vietnam's six cables down at once, affecting roughly 30% of international internet traffic. Each repair typically takes weeks to months.

This article is not about redundancy theory. It answers four specific questions someone responsible for systems needs to answer today: is it really the cable, where does my traffic go, where do my servers sit, and is anything worth buying.

Question 1 — Is it really the cable? Two commands, one minute

Before calling the carrier or blaming anything, separate the domestic leg from the international one. Open a command window — on Windows press the Windows key and type cmd, on a Mac open Terminal — and run two commands:

• ping vnexpress.net — a destination hosted domestically.
• ping 8.8.8.8 — a destination abroad.

Read the time column (time=), measured in milliseconds. How to interpret it:

• Domestic under 30, international jumping above 150 or showing packet loss — the problem is on the international leg. That is the signature of a cable incident.
• Both slow — the fault is inside the office: local network, wireless access point, or that computer. You have just saved a call to the carrier.
• Both fast but one application still slow — the problem is that service itself, not your connection.

These three cases call for three completely different actions, and telling them apart takes a minute.

Question 2 — Where does my traffic leave the country? One command, ten minutes

This is the question most businesses never ask, despite it being answerable with a tool already on every computer.

Run the path-tracing command — tracert on Windows, traceroute on Mac and Linux — against a service you genuinely depend on:

• Windows: tracert drive.google.com
• Mac or Linux: traceroute drive.google.com

The command lists each hop your data passes through. Two things to read:

• Which hop's latency jumps. Typically the first seven or eight hops sit at a few milliseconds (still domestic), then suddenly jump to 40, 70 or 200. That jump is the international exit.
• The name of that hop. Hostnames usually contain three-letter airport codes: hkg is Hong Kong, sin is Singapore, tok or nrt is Tokyo, lax is Los Angeles, sjc is San Jose. That tells you where your traffic leaves — and from there, which cable it most likely uses.

The most valuable way to use this command: run it on *both* lines you are paying for. Plug into line A and run it; switch to line B and run it again. If both exit through the same city, those two lines most likely rest on the same cable — meaning you are paying for two contracts and have one real path.

This is the technical reason behind a very common misunderstanding: the unit of failure is not the carrier, it is the cable. Vietnam has six — AAG, IA, AAE-1, APG, ADC and SJC2 — and every domestic carrier buys capacity across that same set. When SJC2 breaks, it breaks for everyone holding capacity on it, whoever you signed with.

To watch continuously rather than run once, use mtr on Mac and Linux, or WinMTR on Windows — both free. They repeat the trace and show packet loss *per hop*, pointing directly at the congested segment instead of just saying "the network is slow".

Question 3 — Where are my servers? One ping

Many businesses do not know where their systems run, because the technical side chose at build time and nobody asked again. The figures below answer it in ten seconds: run ping against your system's domain and read the time.

• Under 30 milliseconds — hosted domestically. Cable incidents barely touch it.
• Roughly 40 to 90 — Singapore, Hong Kong or Japan. Affected, but tolerably.
• 170 and above — the US or Europe. This group suffers most in every cable incident, because each interaction crosses half the planet and back.

These are normal-condition figures; during an incident they all rise, but their relative order holds, so they still work for classification.

If a system you use daily sits in the third group, this is the thing to do before anything else. Moving a server from the US to Singapore typically cuts latency from around 200 to around 60 milliseconds — three times faster on every interaction, every day, not just during an incident. Most cloud providers offer a Singapore region, and for a small system the move is usually a matter of rebuilding in the new region and migrating the data over a weekend evening.

Two more items belong to this group, both with free options:

• Serve static content from multiple locations. Images, stylesheets and scripts do not need to travel from the origin server every time. Content delivery services with free tiers — Cloudflare being the most widely used — keep copies at many points and serve users from the nearest. For an ordinary business website this removes most of the traffic that would otherwise cross a submarine cable.
• Keep a local copy of anything downloaded repeatedly. If your engineers pull the same software libraries on every rebuild, a caching server in the office eliminates that traffic entirely.

Question 4 — Is anything worth buying, and at what price

This is where money most often goes wrong. The four options below are ordered by real cost, with the condition for choosing each.

Level 0 — mobile data as a fallback. Almost no extra cost. Mobile networks also reach the world over submarine cables, so this is not true path redundancy. But it uses different capacity, is often less congested than fixed lines at peak, and above all it is already there. For an important meeting or an urgent transaction, tethering from a phone is a thirty-second solution. Every business should have this ready before considering the three levels below.

Level 1 — shorten the path. Free to a few hundred thousand VND a month. Exactly the three items above: move servers closer, use a content delivery service, cache locally. This group has the highest return per unit of cost and should be exhausted before buying any connectivity.

Level 2 — an international leased line with a routing-direction request. Roughly 5 to 18.5 million VND per month. Here is what most businesses do not know: with leased line services you can request a preferred routing direction rather than leaving it to chance. Which means "which cable does my backup line use" is not merely a question — it is a negotiable term at signing.

The question to put to the carrier, in writing, should be specific: *which direction does this line prefer, where does traffic move when that direction fails, and can you commit that my second line does not share a direction with my first.* You will not always get a hard commitment, since routing configuration shifts with conditions. But asking alone separates providers who answer concretely from those who answer in generalities.

This level suits businesses where slow time converts directly into money: trading floors, international call centres, teams operating systems abroad. If you cannot convert it, you are not there yet.

Level 3 — satellite internet. Around 10.4 million VND for hardware, monthly fees from 1.7 million. Starlink began serving customers in Vietnam on 13 August 2026. The standard business kit was announced at 11,578,900 VND, and business plans are billed by data volume rather than a flat rate.

Know the conditions before building plans on it: this is a pilot phase capped at 600,000 subscribers nationwide, running five years and ending before 1 January 2031. Satellite is also affected by heavy rain, and while latency is far better than older-generation satellite it remains higher than fibre under normal conditions.

Its value lies elsewhere: a backup path does not need to be as fast as the primary — it needs to fail for different reasons. Satellite is not caught by ships' anchors and not affected by undersea seismic activity. It is the only one of the four levels that genuinely avoids submarine cables, and that is the entire case for it.

Why two lines still fail to switch over

One detail undermines many redundancy investments, and it explains why question 1 matters so much.

Ordinary dual-line routers switch to the backup when the primary dies: signal lost, no response. That is a clear condition and they handle it well.

But a cable break does not kill the line. Traffic shifts onto the remaining cables, so your line keeps working — just far slower. To the equipment every check still passes, so it does not switch, behaving exactly as programmed, while staff sit watching a progress bar.

If you buy a dual-line router, ask exactly one question: does it switch on a dead line, or on latency and packet loss crossing a threshold? The second kind — usually sold as SD-WAN, meaning equipment that measures each line's quality in real time and picks the better one — is what handles this situation. But it only helps when your two lines genuinely take different directions, which the tracert from question 2 will tell you.

Drills: simulate slow, do not unplug

The drill most places run is unplugging the cable. That is a rare situation and the easiest one to handle. The real situation is a slow network, not a dead one.

To simulate on a Linux machine, one command adds 200 milliseconds of latency and 3% packet loss:

• On: sudo tc qdisc add dev eth0 root netem delay 200ms loss 3%
• Off: sudo tc qdisc del dev eth0 root netem

Without a Linux machine, a simpler office-wide version: have a group of staff work a full morning through a phone hotspot in a weak-signal spot. The output is the same — a list of which tasks still work and which stop entirely.

That list decides the budget. Split daily work into two groups:

• Tolerates slowness: sending and receiving email, background file sync, backups, scheduled report downloads. A few extra minutes goes unnoticed.
• Does not tolerate it: video meetings, working in an admin interface hosted abroad, typing commands on a remote server, selling through a system hosted far away.

The second group is usually very short — three or four activities — and it is the only list worth spending money to protect. A great deal goes into raising total bandwidth while the real problem sits in those few latency-sensitive activities that only needed to be moved closer.

What about the longer term

Vietnam's strategy for its submarine cable system to 2030 targets at least 10 new cables, taking the total to at least 15 with a minimum capacity of 350 Tbps, including at least one Vietnamese-owned cable and one international terrestrial route. The terrestrial route matters because it fails for entirely different reasons than a submarine cable.

Two things to hold on to. First, the timeline is years away while incidents happen several times a year. Second, both terrestrial and submarine cables are carrier-level infrastructure — a business does not buy a cable directly, it buys capacity with a commitment about direction from a carrier. So the work does not change: establish where your traffic goes, and request a routing direction when signing a leased line.

Summary: four actions, in order

• Today, one minute: run ping against a domestic destination and an international one, to know where the problem is before calling anyone.
• This week, ten minutes: run tracert on each line and see which city they exit through. If they match, you are paying for two lines and have one.
• This month, one evening: ping your own system. Above 170 milliseconds, move it to Singapore — the single change that removes the most waiting time, with no new connectivity purchase.
• Once those three are done: then consider a leased line with a routing-direction request, or satellite, depending on whether your intolerant-of-slowness group converts into money.

If you want someone to run this check with you, or to set up continuous measurement before you need it, get in touch or see KonexForge's Server & Database capability.