LichunLars

Not the issue I care about but since you asked: 12ms is fine. My old setup was 2ms to a local provider. The extra 10ms is the cost of routing through RackNerd's filtering. Acceptable...

Luxembourg to Bucharest is a well-known path. You were probably hitting Cogent to Frankfurt then back down, or worse. Not BuyVM's fault specifically, but yeah, their single upstream there doesn't have great Balkan peers.

If they...

HostHatch route via CAT Telecom then Singtel adds 8ms vs direct CMI path. Here's Hetzner for comparison:

[code]
traceroute to hetzner.com
1 192.168.1.1 0.5ms
2 core1.lax 1.2ms
3 he.net 4.8ms
4 ntt.sg 158ms <-- bad...

What ping are you actually seeing out of there? Small jurisdictions usually mean one transit provider and a route that goes sightseeing before it reaches anywhere useful

Route matters more than CPU at this tier

[code]
Contabo AMS -> my eyeball ISP:
traceroute to 185.x.x.x
1 192.168.1.1 0.5ms
2 isp-gw 2.1ms
3 ams-ix-peer 4.8ms
4 Contabo.edge 5.2ms

Hetzner NYC -> same:
1 192.168.1.1 0.5ms
2 isp-gw 2.1ms
3 nyiix-peer...

Provider: InterServer
Issue: Their looking glass shows my prefixes via direct peering. Production traffic routes through CloudCone transit, adding 34ms.

Evidence:

[code]
--- looking glass output ---
BGP table entry for 203.0.113.0/24
Paths: (1 available, best #1)
Not advertised to any peer
645...

Quick route check from my end (Copenhagen residential fiber)

[code]
$ ping -c 5 contabo.com
PING contabo.com (185.x.x.x) 56(84) bytes of data.
64 bytes from ...: icmp_seq=1 ttl=56 time=11.2 ms
64 bytes from ...: icmp_seq=2 ttl=56 time=11.4 ms
64 bytes from ...: icmp_seq=3 ttl=56 time=11.1 ms
64 byt...