Skip to content

Anyone getting packet loss to EU?

VPS Hosting by carlos2 24 replies 3.4K views
#1

Been on Contabo Munich for two weeks and routing to EU is cooked. 15% loss to Frankfurt, sometimes worse. Tracert dies in Sofia half the time. Anyone else seeing this? Also which providers take XMR in TR region, need a backup

not your keys, not your coins
#2

My MTR shows consistent loss at a carrier hotel in Sofia, which suggests a cross-border fibre problem. I have opened a ticket requesting transparency on whether this falls under their GDPR Article 32 obligations for network resilience. Here is the relevant excerpt from my logs:

Neuland. Aber schnell.
#3

Yeah I saw this too on my RackNerd box ran mtr for like an hour and it was dying in bulgaria every time I thought it was my config but guess not anyway Contabo takes xmr I think? Not sure about OVHcloud

#4

Contabo takes XMR, OVHcloud does too

#5

Actually, (if you look at (the (bgp.tools) path visualisation)) the OVHcloud prefix is being announced via (a (different (upstream))) than Contabo, yet both converge at (that (Sofia (IX))). So the fault is likely (the (submarine (cable (landing)))) or (a (construction (incident (near (the (border)))))). Helpful to compare:

mtr -r -c 100 target
from both providers to isolate (the (layer (2 (versus (layer (3)))))).

#6

Anyway I stayed up until 4am running packet captures on my KnownHost istanbul box and idk why I do this but the loss pattern looks like clockwork, every 90 seconds, which screams bufferbloat on a failing link. Here's my ugly bash one-liner:

while true; do ping -c 1 8.8.8.8 | grep time; sleep 1; done | tee pinglog.txt
Contabo confirmed cable fault in bulgaria in their status page btw

builds at 3AM, sleeps at noon
#7

90 second interval is brutal, definitely bufferbloat

#8

klaus said:
90 second interval is brutal, definitely bufferbloat
Not necessarily. The 90s cadence aligns with BGP hold-timer expiry on a flapping session. If the Sofia IX route-server is resetting, you'd see synchronized loss across all peers. My NUC lab replicates this with bird2 when I overload the CPU.
watch -n 1 'ip route show | wc -l'
to check prefix churn.

#9

sagepat said:
Not necessarily. The 90s cadence aligns with BGP hold-timer expiry
Interesting. I have updated my ticket to Contabo requesting their BGP session logs for the affected period. Under BNetzA regulations they must retain these for 10 weeks. If they refuse I will escalate to the Bundesnetzagentur.

Neuland. Aber schnell.
2 #10

This.

Post a reply

You need an account to reply. Log in or register to join the conversation.

Post reply Preview Save draft