Latency from South Asia: which server to pick
South Asian routing is dominated by a small number of submarine cable paths and a handful of interconnection points, so traffic between two nearby countries often travels via Singapore or Mumbai. The nearest server on a map is regularly not the fastest one in practice.
4 min read
South Asia is the region where "pick the nearest server" is least reliable advice. Terrestrial interconnection between neighbouring countries is thin, international capacity is concentrated in a few submarine systems, and a great deal of regional traffic reaches its destination by leaving the region first.
Why the routing looks like this
Two reasons, both structural.
Submarine cable geography. The systems serving South Asia land at a small number of points, and the routes run east toward Singapore or west toward the Gulf and Europe rather than between neighbouring countries.
Sparse regional peering. Networks in adjacent countries frequently do not peer directly, so traffic between them transits a third-country exchange. Singapore is the usual one going east, Marseille or the Gulf going west.
The practical result is that a connection from Dhaka to a server in a nearby country may travel several thousand kilometres in each direction.
What to expect
| From Dhaka to | Typical round trip |
|---|---|
| A local Bangladeshi server | 5 to 25 ms |
| Mumbai | 45 to 80 ms |
| Singapore | 60 to 90 ms |
| Dubai | 90 to 130 ms |
| Frankfurt or London | 130 to 180 ms |
| US east coast | 210 to 260 ms |
Those ranges are wide because the variance is genuinely large: which local provider you are on materially changes which international path your traffic takes.
The comparison worth running
Test three exits and keep the best:
- A local server, if one exists and is genuinely local rather than a virtual location. The latency check tells you which: under 30 milliseconds is local, 60 to 90 is Singapore wearing a local label.
- Singapore, which is the region's interconnection hub and frequently beats a nominally closer option.
- Mumbai, which is the strongest option for anyone whose traffic goes west or who needs an Indian origin — with the retention caveat below.
Run each with the speed test, three times, take the median. The difference between the best and worst of the three is often a factor of two.
The India caveat
CERT-In's April 2022 directions require VPN providers serving Indian users to retain subscriber records for five years. Several international providers responded by removing physical hardware from India and offering virtual Indian locations hosted in Singapore.
For a South Asian user choosing Mumbai for latency, that matters twice over: the jurisdiction may not be what the label implies, and the latency may be Singapore's rather than Mumbai's. The same ten-second ping answers both.
Local network conditions
Consumer connections in the region vary enormously, and international bandwidth is often the constraint rather than the last mile. That has two consequences for a VPN.
Peak-hour congestion on international links affects a tunnel more than local traffic, because all tunnel traffic is international by definition. And the tunnel's benefit is larger here than in markets with dense local peering, because more of your traffic was crossing an international path anyway.
Practical settings
MTU is worth setting explicitly. Long submarine paths with multiple carrier tunnels in between make path MTU problems more common than on short terrestrial routes, and 1380 is a safe starting point.
Keepalive matters on mobile, which is the majority of connectivity in much of the region, because carrier NAT timeouts are short.
The Bangladesh and Singapore pages cover what each location is and what the local network environment looks like.
Reading a traceroute in this region
A traceroute is unusually informative here, because the hops name the path and the path is the whole story.
traceroute vpn-endpoint.example.net
Look for the point where latency jumps by 40 milliseconds or more in a single step. That is the submarine leg, and where it appears tells you whether your traffic is going east or west. Hostnames on the intermediate hops frequently name the city, which is the fastest way to learn what your provider's international routing actually does.
Doing this once, on your own connection, is worth more than any general table — including the one above — because the variance between providers in the same city is large.
Cable faults are a normal condition
Submarine systems serving the region are repaired regularly, and a fault on a major system reroutes enormous volumes of traffic onto the remaining paths. The effect is visible: latency rises, peak-hour congestion worsens, and it persists for days or weeks until a repair ship completes the work.
If performance degrades suddenly across every server and nothing local changed, this is worth ruling out before troubleshooting anything on your own side. Local network operator announcements usually mention it.
What a tunnel changes here
More than in a well-peered market. Where a large share of your traffic is already crossing an international path, the tunnel adds proportionally less overhead, and routing through a well-connected exchange can occasionally improve the path rather than lengthen it.
That is not a general claim that a VPN makes things faster. It is a specific observation that the usual ten to thirty percent cost is measured against a baseline that is already long, and the relative penalty is therefore smaller than someone in Frankfurt would experience.
