The misconception we hear most: "two data centers means two networks". Two IP plans, a VPN tunnel in between, and an application that has to know which site it runs at. You can build it that way, but you don't have to. On our network, dedicated servers in Ede, Naaldwijk and Amsterdam can share one private segment, so they see each other as if they were in the same rack.
Here's how that works, what customers use it for, and when we'd advise you not to do it.
Why spread servers over two sites
One data center is one place where things can go wrong: a power failure, a fire, a digger cutting through a cable. A second site takes that risk out of your setup. What you do with that second site varies:
- Failover — a standby server that takes over when the first site drops out.
- Replication — your database or storage writes continuously to a second site, so a failure doesn't cost you data.
- A cluster across two sites — for example a virtualisation cluster that live-migrates virtual machines between locations without re-addressing them.
For those last two, the software usually wants the machines on the same network. That's when you need Layer 2 between the sites.
How it works: your own segment on our fabric
Our locations are linked over our own lit dark fiber. On top of that runs an EVPN-VXLAN fabric: an open standard that carries Ethernet across a routed network, without the weak spots of a classic stretched VLAN.
What that gets you:
- Your own isolated segment — your traffic is strictly separated from other customers'.
- All paths active — no spanning tree blocking half your links; if a path fails, the fabric routes around it.
- Plain Ethernet on your side — you keep your own VLANs, your own IP plan and your own routing.
If you want to go deeper into the technology, our sister company Xyphen IT explains the control plane and how it differs from classic Layer 2 in EVPN-VXLAN explained. How the fabric sits under all our services is covered in our infrastructure article.
VRRP or an anycast gateway
A segment across two sites needs a gateway, and we pick one of two implementations with you. VRRP gives you one virtual gateway address that moves to the other side during a failure; simple and predictable. An anycast gateway in the fabric is active at every site at once, so your servers always use the nearest gateway and traffic doesn't have to travel to the other city first. Which one fits depends on your setup; neither is the "legacy" option.
Two things to think about up front
Distance takes time. A packet between Naaldwijk and Amsterdam travels further than one inside a single rack. For failover and asynchronous replication you won't notice; with synchronous replication every write waits for the other site. Test that with your own workload before you build a production design on it.
Two is an awkward number for clusters. If the link between two sites drops, each side assumes the other has failed. Without a referee you get split brain: two sides that both act as primary. The fix is a third vote, a witness or quorum node, preferably at a third site. With three locations on our network, that one is simply available.
Honest advice: route where you can
A broadcast domain spanning two cities deserves a deliberate design. EVPN-VXLAN removes the biggest risks of stretched Layer 2, but it's still wise to stretch only what genuinely needs it: live migration, clusters that demand L2 adjacency. A web server at one site and a database at the other can talk over a routed link just fine. If we conclude together that you don't need Layer 2, that's a good outcome too.
How to get it
It's custom work, so it starts with a conversation about what you want to achieve. Existing dedicated servers in Ede and Naaldwijk can be placed in one segment; a dedicated server in Amsterdam is possible too, with a longer lead time. Combine that with your own hardware in colocation and it can join the same segment. And if you want to connect straight to an internet exchange from your segment, the same fabric makes that possible.
See our data centers and network or get in touch; you'll talk to an engineer who knows the fabric first-hand.
Frequently asked questions
Can I use my own IP addresses and VLANs in the segment?
Yes. The segment hands off plain Ethernet. Your IP plan, the VLANs inside your segment and your routing stay yours.
Do I need a third site?
Not for failover or replication. If you run a cluster that relies on quorum, we recommend a witness at a third site, so a failing link between two sites can't cause split brain.
Can I put VPSes and colocation in the same segment?
Yes. Dedicated servers, VPSes and your own hardware in colocation can share one private segment, across multiple sites.