Server location: Germany, the USA or France
Pick the location by the geography of your audience: the facility should sit as close as possible to the country most requests come from. After that the deciders are jurisdiction, the type of DDoS protection, the traffic allowance and the price. ZevsHost runs three facilities — Germany (Dusseldorf), the USA and France (Strasbourg). You can verify your choice in ten minutes with ping and traceroute from machines in the target region.
- Audience in Europe means Germany or France; audience in North and South America means the USA.
- Eyeballing does not work: measure RTT and the route from several points inside the target region to a test IP of the facility.
- Extra milliseconds of RTT are multiplied by the number of round-trips: the TLS handshake spends them several times before the first byte.
- Germany and the USA provide L3/L4 DDoS protection, with L7 on Pro and above; France runs Anti-DDoS Arbor.
- Only Dedicated Start US has a traffic allowance: 1 Gbit/s and 30 TB per month; every other plan is unmetered.
What actually sets the response time
Latency is the sum of physical distance, the number of transit hops and the quality of peering. Distance sets a floor that no CPU and no disk will reduce; everything above it depends on which networks a particular carrier hands your traffic to.
The same facility answers differently for two users in the same city: they have different ISPs and different exchange points. Measure from the networks your audience actually uses, not from your work laptop over a corporate VPN.
Latency weighs most where a single page fires many sequential requests: REST APIs with dozens of calls, game protocols over WebSocket, admin panels. For static files and video, port width and caching matter more, which is covered separately in the article on the server uplink and traffic limits.
How to measure the latency yourself
Take a test IP at the facility you are considering and run three tools against it from a VPS or a server in the region where your users live, not from home Wi-Fi.
# Debian/Ubuntu: measurement tools
apt-get update && apt-get install -y iputils-ping mtr-tiny traceroute curl
# RHEL/AlmaLinux/Rocky
dnf install -y iputils mtr traceroute curl
# 100 packets at 0.2 s intervals: read min/avg/max and packet loss
ping -c 100 -i 0.2 TEST_IP
# per-hop route with loss, report written to a file
mtr -rwzbc 200 TEST_IP > mtr-de.txt 2>&1
# where the time grows: TCP traceroute to port 443
traceroute -T -p 443 -n TEST_IPRead the output like this: avg from ping is your baseline for comparing facilities, and mdev is jitter, which matters more than the average for games and voice. Loss reported by mtr on the final hop means a problem at the facility; loss on an intermediate hop that does not repeat further down the path is ICMP rate limiting on a transit router and is ignored.
For web workloads it is more useful to measure full HTTP timing rather than ICMP, because it includes the TCP handshake and TLS. Run the request a dozen times in a row so that you see the spread instead of a single value.
# full request timing: DNS, connect, TLS, first byte
curl -o /dev/null -s -w 'dns=%{time_namelookup} conn=%{time_connect} tls=%{time_appconnect} ttfb=%{time_starttransfer} total=%{time_total}\n' https://TEST_HOST/If the route looks odd — loops, a sharp jump at a network border, asymmetry — a step-by-step breakdown is in the article on Linux network diagnostics.
Facilities compared
| Parameter | Germany (Dusseldorf) | USA | France (Strasbourg) |
|---|---|---|---|
| Closest to | Central and Eastern Europe | North and South America | Western and Southern Europe |
| Entry plan | Start DE, $49 | Start US, $59 | Start FR, $55 |
| Entry plan uplink | 100 Mbit/s unlimited | 1 Gbit/s, 30 TB | 500 Mbit/s unlimited |
| Top plan | Pro DE, $99 | Enterprise US, $149 | Pro FR, $95 |
| DDoS protection | L3/L4, plus L7 on Pro | L3/L4, plus L7 on Pro and above | Anti-DDoS Arbor |
| Dual-socket option | none | Enterprise US: 2x E5-2680v4, 64 GB, 4x1 TB NVMe RAID10 | none |
Germany is the cheapest entry point, but Start DE has a 100 Mbit/s port. A dual-socket machine with out-of-band KVM/IPMI included exists only in the USA. All configurations are listed on the dedicated servers page.
Criteria beyond latency
Traffic and port width
Unmetered in Germany and France removes the allowance question but pins you to the port width: a fully saturated 100 Mbit/s port moves roughly 1 TB per day. Start US has a gigabit port, but with a 30 TB monthly counter.
Protection against attacks
Germany and the USA filter L3/L4 on every plan, and L7 is available from Pro upwards. France works through Anti-DDoS Arbor. What is dropped at each layer is covered in the article on DDoS protection for a dedicated server.
Addresses and out-of-band management
An extra IPv4 costs $5/month at any facility, KVM/IPMI access is also $5/month, and on Enterprise US it is included. If you plan virtualization, count the addresses in advance: see extra IPv4 addresses and subnets.
Changing the location after launch means changing the IP address. Lower the TTL of your DNS records in advance, a day or more ahead, or part of your users will keep hitting a powered-off machine for hours. Verify the move from the logs of the old server rather than by checking that the site opens: while requests keep arriving in its access log, the migration is not finished. The second trap is hard-coded IP addresses in configs, firewall rules and payment gateway allow lists; grep for them before the switch, not after.
Key takeaways
- Audience geography decides the location; verify it with measurements taken from machines in the target region, not from your own desk.
- ping gives baseline RTT and jitter, mtr shows the hop where loss appears, and curl with -w gives real time to first byte.
- Germany is the cheap entry point, the USA has the dual-socket configuration and IPMI included, France offers Arbor and unmetered 500 Mbit/s on the entry plan.
- Before switching facilities, lower your DNS TTL and strip hard-coded IP addresses out of your configurations.