Internet Censorship, VPNs, and Why an IP Can Suddenly Stop Working

An IP address works perfectly from Amsterdam, Sofia, and Singapore, but not from one mobile network in one country. The server is online, the routes look normal, and monitoring is green. Naturally, the first support ticket says: “Your network is down.”

Sometimes it is. Sometimes the missing packet has encountered something more political than a failed switch.

Internet censorship affects websites, messaging platforms, VPNs, cloud applications, game servers, media, and ordinary business systems that happen to share infrastructure with a blocked service. It also changes quickly. An address reachable yesterday may fail today and return next week without its owner changing anything.

This article covers only the technical aspects of using cloud services from regions where access to external internet resources may be restricted, filtered, slowed, or otherwise disrupted. It explains how those restrictions can affect connectivity, testing, support, and service availability.

We do not evaluate the political or social circumstances behind these measures. We also do not analyse local legislation, decide whether a restriction is lawful or justified, or provide legal advice. Customers are responsible for understanding the rules that apply in their location. This article does not provide instructions for bypassing access restrictions.

Internet filtering grew up with the public internet

Governments often controlled print, broadcasting, and telecommunications long before TCP/IP arrived, although the degree of control varied enormously between countries. Many still do. Network-level internet filtering became practical in the mid-to-late 1990s, when public access expanded but international connections still passed through a small number of gateways.

China began regulating international computer networks in 1996 and started systematically blocking some foreign websites that year. The phrase “Great Firewall of China” appeared in a 1997 WIRED article. Early filtering commonly relied on lists of IP addresses, domains, or URLs. Saudi Arabia routed public internet access through centrally controlled filtering systems by the end of the decade, while Iran developed one of the most sophisticated state-mandated systems documented by the OpenNet Initiative in 2004 and 2005.

These early systems were blunt but effective. Block an address at a national gateway and users behind participating ISPs can no longer reach it. The collateral damage was acceptable to the filter, if not to the unrelated websites sharing that address.

Where censorship is widespread

There is no permanent league table because restrictions change during elections, protests, conflicts, security events, and even school examinations. The last example is not satire. Access Now recorded 11 exam-related shutdowns across six countries in 2025, while the Internet Society has documented the practice in countries including Sudan, Jordan, Iraq, Algeria, Syria, and India.

Countries also censor different things. One blocks political media, another gambling or adult content, another messaging applications, and another disconnects mobile internet in selected regions because somebody might share tomorrow’s algebra questions.

Freedom on the Net 2025 assessed China and Myanmar as the worst environments among the 72 countries it covered. It also described extensive or intensifying controls in Iran and Russia. Long-running, large-scale filtering is documented in countries including China, Iran, Russia, Myanmar, Turkmenistan, Vietnam, Pakistan, Saudi Arabia, and several other states in the Middle East and Central Asia. The scope and legal basis differ, so “censored country” is not one technical configuration.

Targeted shutdowns are even more widespread. Access Now documented 313 internet shutdowns during 2025, the highest annual total since its tracking began in 2016. A shutdown may cover a whole country, one province, mobile data, particular platforms, or selected hours. From a user’s desk, all of them initially look like the same deeply informative error: connection timed out.

How modern network censorship works

An operator can filter traffic at several layers, and national systems often combine them.

DNS tampering returns a false address, an error, or no useful answer when a device asks where a domain is located. Changing resolvers may reveal that DNS interference exists, but it does not help when another layer independently blocks the destination.

IP or network blocking drops traffic to a particular IPv4 or IPv6 address, prefix, autonomous system, or route. It is simple and cheap, but shared hosting and cloud networks make it imprecise. Blocking one address can take down many unrelated domains; blocking a larger prefix can make a considerable piece of the internet disappear.

HTTP and hostname filtering can inspect unencrypted web requests. With HTTPS, a filtering device may still observe metadata exposed during connection setup, including the Server Name Indication in TLS configurations where it is not encrypted. This can allow one hostname to be blocked while other sites on the same address remain available.

Deep packet inspection, or DPI, classifies traffic by more than its destination. It can look for protocol signatures, handshake structure, packet sizes, timing, connection patterns, or other fingerprints. Encryption protects content, but encrypted protocols still have observable behaviour. A censor may identify and throttle or reset a connection without reading the webpage, message, or file inside it.

Active probing goes one step further. When traffic resembles a proxy or VPN, a separate system may connect to the suspected server and test how it responds. Research from Princeton documented this cat-and-mouse technique against Tor-related infrastructure in China.

Throttling and selective failure are harder to prove than a block page. A regulator or ISP can make a service extremely slow, reset only some connections, block UDP while leaving TCP, or interfere only on certain networks and at certain times. The service technically remains “online,” in roughly the same way that a restaurant remains open when its front door is welded shut but the lights are on.

At the far end is a partial or complete shutdown. Authorities can order providers to withdraw routes, disable mobile data, isolate international gateways, or permit only approved domestic services. No VPN protocol can repair a physical or routing-level absence of connectivity to the outside world.

The OONI project collects open measurements of DNS tampering, TCP/IP interference, block pages, messaging-app blocking, and circumvention-tool reachability. Its data is useful precisely because censorship may vary between two ISPs in the same city.

What a cloud provider can and cannot control

ITLDC provides cloud computing, servers, IP connectivity, and datacenter infrastructure. We monitor our network, maintain routing, investigate abuse, and fix faults within systems we operate. We do not control a foreign ISP, a national filtering platform, or the rules applied at another country’s border gateways.

A VDS or VPS is a general-purpose virtual server, not automatically a VPN. Customers choose the operating system, software, configuration, and permitted workload. It might host a website, database, development environment, private network, or something else entirely. ITLDC supplies and supports the underlying cloud infrastructure; we do not assign a political purpose to a server simply because VPN software could be installed on it.

If an ITLDC address is reachable globally and traffic leaves our network normally, we may have no technical switch that can make a remote operator accept it. Announcing the same route more enthusiastically is not a routing protocol feature.

We also cannot reliably say why a country or local ISP blocked an address. The trigger might be a specific service, user activity, historical address reputation, a shared network range, protocol detection, an automated false positive, or a government order that is not public. A route failing from one provider is evidence of a reachability problem, not proof of who ordered it or why.

For the same reason, we cannot recommend a particular censorship-circumvention solution. Our specialization is cloud infrastructure, not local freedom-of-expression law, personal threat assessment, or access policy in every jurisdiction. A tool appropriate for a company connecting two offices may be ineffective, restricted, or risky for an individual elsewhere. Customers must evaluate local law and their own situation.

Test before ordering

If reachability from a particular country matters, test it before purchasing a service. Our Looking Glass and reference IP list provides IPv4 and IPv6 addresses for each available location, together with ping, traceroute, and DNS diagnostics.

Test the reference IP for the exact location from the country and networks that matter to you. A successful result from one home ISP does not guarantee the same result on every mobile operator. Where possible, check fixed broadband and mobile connectivity, IPv4 and IPv6, and more than one local provider.

A successful ping only proves basic reachability at that moment. It does not prove that a VPN handshake, UDP transport, application port, or sustained encrypted session will pass the same filter. Where protocol-level censorship exists, no provider can guarantee that a particular technology will work with 100 percent certainty. The route can be open while the application is still detected and interrupted.

What to do when an IP stops working

Start with a few simple checks:

1. Try the service through another ISP or from another country. You can also check it from multiple global monitoring points with Check-Host or ping.pe. If the server responds from several countries but not from your network, the problem is probably regional.
2. Compare your IP with the reference IP for the same location. If both fail from the same region, the problem is probably larger than one server.
3. Try another suitable protocol or technology. A filter may recognise one type of traffic but not another. ITLDC cannot choose or configure a circumvention tool for you.
4. If needed, order a service in another location after testing its reference IP. A different IP may help, but filtering can change again.

When contacting support, include the affected country, ISP, server IP, protocol, approximate time the problem began, and any traceroute results. We can check the server and our network, but we cannot remove a filter operated by another country or ISP.

For important workloads, keep two or three small virtual servers in different tested locations. One server is convenient; several are a continuity plan.

If more than two weeks remain in the paid term, contact support. We can review the situation and may offer a partial refund or transfer part of the remaining paid period to another service. The available option depends on the circumstances and is not automatic. If fewer than two weeks remain, the normal approach is to disable automatic renewal and let the service expire.

Regional blocking does not mean the server is defective. It can also leave us with a technically healthy IP that is difficult to reuse. If you request cancellation, back up all required data first: once the cancellation is processed, the service and the data stored on it may no longer be available.

A fragmented internet costs everyone

Network censorship does more than hide a webpage. It makes legitimate services unreliable, encourages repeated IP changes, increases costs, breaks shared infrastructure, and makes troubleshooting less transparent. Businesses lose access to tools and customers, researchers lose sources, families lose communication channels, and providers spend time diagnosing packets discarded by systems they cannot see.

VPN popularity is therefore both a response to filtering and a reason filters become more sophisticated. The cycle will continue as long as networks are asked to decide not merely where packets should go, but whether users should be allowed to send them.

Our role is narrower and practical: keep our infrastructure working, publish test endpoints, provide honest diagnostics, and explain the limits of what we control. Before ordering, test the location. If access later disappears and the evidence points to regional censorship, the practical next step is to accept that limitation and decide whether another tested location or service makes sense.

Support can confirm what is happening on our side, but repeated tickets or urgent demands cannot make a foreign ISP remove its filter. We understand that losing access is frustrating, and we will review a documented case, including possible cancellation when substantial prepaid time remains. Regional reachability and the reputation of an IP used by the customer, however, remain the customer’s responsibility. Hope is pleasant, but it has never completed a traceroute, and neither has an ultimatum sent to the wrong network operator.

Sources and further reading

• Freedom House: Freedom on the Net 2025
• Freedom House: anti-censorship tools and the open internet
• Access Now: 2025 internet shutdowns report
• OONI: internet censorship measurements and methodology
• Berkman Klein Center: early internet filtering in China
• Berkman Klein Center: early internet filtering in Iran
• ITLDC Looking Glass and location reference IPs