Protocol comparison
| Protocol | Speed | Audit status | Best for |
|---|---|---|---|
| WireGuard | Fastest (~4k lines of code) | Multiple audits, kernel-integrated on Linux | Everything — default choice now |
| OpenVPN | Slower but stable | Battle-tested 20+ years | Compatibility, obfuscated setups |
| IKEv2/IPSec | Fast, mobile-friendly | Solid but closed implementations vary | Mobile devices reconnecting networks |
| Proprietary (NordLynx, Lightway…) | Very fast | WireGuard-based, open-source clients | Fine if code is auditable |
| PPTP/L2TP | — | Broken/weak | Avoid — deprecated |
Settings that actually matter
- Kill switch — blocks traffic if the tunnel drops. Verify it works: disconnect mid-download and check for leaks.
- Split tunneling — route some apps outside the VPN (banking sites that block VPN IPs) while keeping the rest inside.
- DNS handling — ensure the app forces its own DNS; third-party DNS leaks queries to your ISP.
- IPv6 handling — many leaks come through IPv6 when the app only tunnels IPv4.
- Auto-connect on untrusted networks — essential for laptop/phone use.
Manual config vs provider apps
| Provider app | Manual config | |
|---|---|---|
| Setup effort | Zero | Moderate |
| Kill switch/leak protection | Built-in | Your responsibility |
| Transparency | Trust the app | You control everything |
| Key rotation, updates | Automatic | Manual |
| Who it's for | Most users | Advanced users, routers, custom setups |
How to verify your VPN is actually working
- ipleak.net — check IP, DNS servers, WebRTC leaks.
- dnsleaktest.com — extended test for DNS leaks.
- Kill-switch test — download a large file, disconnect VPN, watch traffic die.
tracert/tracerouteto a known host — confirm the first hop is the VPN, not your ISP.
Sources
VPN software — the protocol wars and what survived
The software layer tells the real story. OpenVPN (2001) won two decades on configurability — but its ~100k+ lines made audits expensive and throughput modest. WireGuard (announced 2015, kernel-merged 2020) rebuilt the stack around ~4,000 auditable lines, modern crypto and near-instant handshakes; provider implementations (NordLynx, Lightway) filled its static-IP gaps.
The settings that actually protect you are fewer than apps suggest. The kill switch is the only non-negotiable — everything else is preference. App vs manual config is a trust question: provider apps add leak protection but also their own telemetry; manual WireGuard configs remove the provider's software entirely.
Counted and measured
| Item | Reality |
|---|---|
| Protocol code auditability | WireGuard ~4k LOC vs OpenVPN ~100k+ vs IKEv2 closed stacks |
| Real-world throughput delta | WireGuard typically 2-4× OpenVPN on same link |
| Kill switch failure modes | most leaks happen on reconnect/roaming, not steady state |
| App telemetry | most provider apps phone home usage stats — check the privacy policy for 'diagnostics' |
| DNS leak rate | most common real-world failure — force provider DNS or verified resolvers |
The settings checklist that matters: kill switch on, protocol = WireGuard unless a network blocks UDP, DNS forced through the tunnel, app auto-start off on untrusted machines. That covers ~95% of real-world leak scenarios; the rest is theatre.
What the install base teaches
The download numbers across platforms show where the real usage is: mobile clients now account for the majority of installs, but desktop remains where the protocol engineering shows — WireGuard's 4,000-line kernel implementation versus OpenVPN's sprawling codebase is the canonical example of how design simplicity became a security feature, not just a performance one.
Choosing between the generations
The practical 2026 matrix: WireGuard-class protocols (including NordLynx, Lightway and friends) for anything latency-sensitive; OpenVPN remains the fallback where network filtering is hostile — it survives places WireGuard doesn't because it can masquerade as ordinary TLS. Protocol choice now matters more than provider branding in most use cases.
The install-and-forget test that separates software tiers: does the client reconnect cleanly after sleep/network-switch, does the kill switch survive edge cases, and can you read the source or the audit? Everything else is skin.
Expert context · 2026
“The protocol matters less than the implementation — a misconfigured WireGuard leaks less than a misconfigured OpenVPN only because there's less to misconfigure.”
— Jon Callas, cryptographer, EFF · eff.org