Contabo "gaming optimized" KVM has 150-300ms game latency despite clean traceroute

Posting diagnostics because I'm stumped. Running a game server on a Contabo KVM in their "gaming optimized" tier.

Symptoms: Players report 150-300ms in-game latency. Server-side metrics show 5ms frame times.

Traceroute from my home to server:

 1  192.168.1.1    0.5ms
 2  10.0.0.1       2.1ms
 3  core.isp.net   5.3ms
 4  peer1.isp.net  8.7ms
 5  Contabo.edge 12.4ms
 6  game.Contabo 12.6ms

Looks clean. 12.6ms. Beautiful.

But: `mtr` shows 0.5% packet loss at hop 5 only under UDP flood. TCP clean.

Server side: `tc -s qdisc show dev eth0` reveals Cake with 100ms bandwidth limit, but I'm nowhere near bandwidth cap. `cpulimit` in cgroup? No. Steal time under 1%.

What I've ruled out:

• Virtualization tax — it's KVM, not OpenVZ garbage
• Bufferbloat on my end — tested with flent, A+
• Game server CPU — headroom everywhere
• Symmetric routing — return path similar

Something between "traceroute perfect" and "game feels awful." What am I missing?

¡¿12.6ms traceroute but game is garbage?! Is very strange, my friend. Jajaja I had this same problem with RackNerd once, the routing was "perfect" but the UDP was going through a different path than TCP, ¿you know? The traceroute use ICMP and the game use UDP and sometimes the ISP treat them different... Jajaja I wrote a long explanation but deleted it, too much words. Check your provider's QoS for UDP ports, is very common they deprioritize game traffic. ¡good luck!

— Routing Asymmetry Check

@kate3 Good structured report. One addition: run concurrent traceroutes in both directions simultaneously. Contabo's "gaming optimized" network uses source-based routing for UDP/3074 (Xbox) and UDP/3478 (PlayStation). If your game server runs on a non-standard port, it may fall through to best-effort transit.

We've seen this with three other customers this quarter. The "optimization" is port-specific, not protocol-specific.

Worth testing: temporarily move your server to UDP/3074 and measure. Not a fix, but confirms the hypothesis.

— Admin

SNAPPED UP a Contabo €4/mo deal last week for my Valheim server and SAME ISSUE. traceroute clean, game unplayable. Their "gaming" routing only activates on UDP/27015-27050 (Steam)!!! Moved my server to 27016 and BAM 15ms stable.

Your provider might have similar whitelist? Check the fine print or just... test ports.

@kate3 Your mtr UDP loss at hop 5 is the clue. Contabo peers with Cogent at that edge; return path likely differs. Request a looking glass from their support, or use their LG if public.

More specifically: their "gaming optimized" product likely uses BGP communities to steer UDP toward a specific transit provider with lower latency but higher per-Mbps cost. The community match is almost certainly destination-port-based, as Admin noted. If your game uses a non-standard port, the route doesn't get tagged, and you ride the cheap transit with bufferbloat and occasional loss.

You can verify with `traceroute -U -p 3074` versus `traceroute -U -p [your port]` from an external host. Different AS paths would confirm.

The real fix is getting Contabo to document their community strings, or moving to a provider that doesn't do application-specific routing without disclosure. This is technically not false advertising but it's close.