Looking Glass

Check the route, latency and throughput between our region, eu-east-ro-1 in Bucharest, and wherever your users are, before you create an account. The test host sits on the same uplink as customer machines.

From us to you

The Looking Glass runs ping, mtr and traceroute from our side to any public address, over IPv4 or IPv6. It is the quickest way to hand your provider a trace in the direction they cannot see.

Download test files

100 MB · 1 GB · 10 GB

The files are random bytes, so compression anywhere on the path cannot inflate the result. Writing to /dev/null measures the network and leaves your disk out of it.

wget -O /dev/null https://lg.serverscamp.com/files/1GB.bin

# pick the protocol
curl -4 -o /dev/null https://lg.serverscamp.com/files/1GB.bin
curl -6 -o /dev/null https://lg.serverscamp.com/files/1GB.bin

iperf3

# from you to us
iperf3 -c lg.serverscamp.com -P 4

# from us to you
iperf3 -c lg.serverscamp.com -P 4 -R

# the same over IPv6
iperf3 -6 -c lg.serverscamp.com -P 4 -R

Ports 5201 to 5205 each run one test at a time. If 5201 answers that the server is busy, add -p 5202 and move up. Tests run over TCP and last up to 30 seconds. A single stream rarely fills a long path, which is why the commands use -P 4.

Route and latency with mtr, traceroute and ping

mtr -rwzbc 100 lg.serverscamp.com
mtr -6 -rwzbc 100 lg.serverscamp.com

traceroute lg.serverscamp.com
ping -c 20 lg.serverscamp.com

In the mtr line, -r prints a report instead of the live screen, -w keeps full hostnames, -z adds AS numbers, -b shows names and addresses together and -c 100 sends 100 probes per hop. On Windows, pathping lg.serverscamp.com or WinMTR gives the same picture.

When you read the report, loss on one hop in the middle that does not continue to the next hops is a router answering probes addressed to itself at low priority, and your traffic passes through it untouched. Only loss that holds all the way to the last line affects a real connection.

What a VM measures

A VM's port is multi-gigabit whatever the plan, and the smallest machine gets the same network as the largest. What you actually see depends mostly on the other end and on latency. The table below is YABS iperf3 output from a ServersCamp VM in September 2026.

Over IPv6 the same run reached 8.46 Gbit/s from Amsterdam. Full output and disk numbers are on the benchmarks page. To repeat only the network part yourself, run curl -sL https://yabs.sh | bash -s -- -fg.

Fair use of the test host

Everyone who checks us shares this one machine. Looking Glass commands are limited to a few per minute per address, iperf3 to 20 connections per address, and a script that pulls the test files over and over is blocked for a day. A normal round of tests never comes close.