Choosing a Dedicated Connection: Inside the Data Center, Between Data Centers, and to the Cloud

8 min readPacketStream

Where the public internet and a VPN are enough, where a dedicated circuit is needed, and how links inside a data center, between sites and to the cloud compare.

On this page
  1. Where a VPN is enough
  2. Which endpoints are you connecting?
  3. Cross Connect: L1 link
  4. Metro Connect: Layer 2
  5. Cloud Connect: to the cloud
  6. Redundancy in paths
  7. What sets the price
  8. Choosing a connection

Link the servers in your colocation rack to a database in the public cloud over a VPN and, most of the time, it simply works. Then replication lag starts to widen only at certain hours, or bulk transfer speeds change from one day to the next, and adding bandwidth does not always fix it. Inside one data center, between data centers and out to the public cloud: what kind of connection does each of these three stretches call for?

Where the public internet and a VPN are enough

The public internet and a VPN are enough where traffic is light and latency is allowed to wobble a little. Development and test environments, management access and small backups on a long cycle all fall here. The trouble starts on stretches where volume grows and response times have to stay steady. Two responses look obvious at that point.

  1. Buy more internet bandwidth → Only your own segment gets wider. The other providers' networks in the middle stay as they are.
  2. Add VPN appliances or tunnels → Encryption throughput goes up, but the tunnels still run over the public internet.

Neither solves the path problem. AWS describes Site-to-Site VPN as an IPsec connection created over the internet. Microsoft explains that ExpressRoute, its own private connection service, offers more consistent latency than typical internet connections because it does not go over the public internet.

Public internet and VPN

  • An encrypted tunnel built over the internet
  • The path crosses several providers' networks, so congestion is out of your hands
  • Fits stretches with light traffic where latency can wobble a little

Dedicated circuit

  • Connects the two endpoints without crossing the public internet
  • Can reduce latency variation caused by congestion on the public internet
  • Fits stretches with heavy volume where response times must stay steady

Think of ordinary roads and a dedicated lane. On an ordinary road your arrival time depends on the traffic around you, however good your car is. On the public internet, packets likewise cross several providers' networks in turn, and you have no control over congestion on those segments. How to measure latency from Korean carrier networks to your server is covered in the PacketStream Pulse article.

Once you have decided you need a dedicated circuit, the next question is which kind.

Which endpoints are you connecting?

Choose the connection based on what you need to connect. Use Cross Connect for a physical link within one data center or Metro Connect to link two data centers in the Seoul metropolitan area. If one endpoint is a public cloud, Cloud Connect extends the connection to the cloud provider’s point of presence.

CriterionCross ConnectMetro ConnectCloud Connect
StretchInside one data centerBetween data centers in the Seoul metro areaFrom a rack in the data center to a public cloud
LayerL1 physical linkDedicated L2 (Ethernet) on a DWDM backboneL2 or L3 interconnection
Bandwidth1G to 400G, by portUp to 100 Gbps per link100 Mbps to 10 Gbps
RedundancySingle or dual pathUnprotected or ProtectedSingle or dual path
What sets the priceData center, number of connectionsThe two data centers, bandwidth, redundancyCloud provider, data center, bandwidth, redundancy

The layer tells you how much of the connection the provider takes on. L1 is the cable itself, L2 is the delivery of Ethernet frames and L3 is IP routing. The higher the layer, the more the provider handles; the lower the layer, the more you decide yourself. Start with a connection inside one data center.

Cross Connect is an L1 link that joins two pieces of equipment in the same data center directly with a cable. From your rack you can reach a carrier or an IX, another customer's equipment in the same building, or another rack of your own. No routing equipment sits in between, so of the three connections this path is the simplest.

Three cable media are supported, and the media on offer differ by data center.

Cross Connect cable media

  • SMF fiber
  • MMF fiber
  • Copper (Cat6)

If physical separation matters for your security requirements, also look at a separate patch panel, armored cable on request and a secured meet-me room (MMR).

Metro Connect: dedicated Layer 2 between data centers

Metro Connect links data centers across the Seoul metro area over dedicated L2 (Ethernet). Equipment in two separate data centers behaves as if it were plugged into one switch, so you keep running your IP addressing and routing the way you already do. Traffic between the data centers stays off the public internet.

Transport runs on a DWDM (Dense Wavelength Division Multiplexing) backbone. DWDM carries several wavelengths on a single fiber and uses each wavelength as an independent channel. You can choose Point-to-Point, which joins two sites, or Multipoint, where several sites share one L2 network. Bandwidth per link is one of the four options below, and you can scale beyond that with LAG (Link Aggregation).

Metro Connect bandwidth per link

  • 1 Gbps
  • 10 Gbps
  • 40 Gbps
  • 100 Gbps

Metro Connect is mainly used in three setups.

  • Backbone between data centers: joins the networks of two data centers into one.
  • Separating production and disaster recovery (DR): keeps the production site and the DR site in different data centers.
  • Data replication: replicates data between data centers.

Reaching a public cloud from here takes one more step.

Cloud Connect: a dedicated circuit to the public cloud

Cloud Connect links your colocation rack in the data center to a public cloud over a dedicated circuit. Each cloud provider has its own private connection service, and Cloud Connect supports the four below.

Private connection services supported by Cloud Connect

  • AWS Direct Connect
  • Azure ExpressRoute
  • Google Cloud Interconnect
  • Oracle FastConnect

AWS explains that Direct Connect lets you connect to AWS services while bypassing internet service providers in your network path.

Port bandwidth is 100 Mbps, 1 Gbps or 10 Gbps, on a dedicated or a shared port. Both L2 and L3 interconnection are supported. Cloud providers' private connection services exchange routes over BGP. AWS lists 802.1Q VLAN and BGP support among the requirements for Direct Connect, and Microsoft likewise explains that ExpressRoute exchanges routes over BGP. With an L2 handoff you run the routing toward the cloud on your own equipment, so whether your team can operate BGP becomes the deciding factor.

For all three connections, if there is only one path, everything stops when that path is cut.

Redundancy is counted in paths

Redundancy is measured not by the number of circuits but by the number of independent paths. Two circuits that run through the same duct can be cut by a single accident. All three connections let you choose between a single path and a dual path, and the choice determines what happens during a failure.

Single path

  • Unprotected on Metro Connect, single path on Cloud Connect
  • If the path is cut, traffic stops until it is restored
  • Fiber recovery on Metro Connect is completed within 6 hours
  • Fits backup transfers that can run a few hours late

Dual path

  • Protected on Metro Connect, Redundant Link on Cloud Connect
  • On a failure, traffic fails over to the second path automatically
  • Fits stretches that carry real-time replication or service traffic

The test is how long your service can hold out with this connection down.

The level of redundancy is set by the weakest segment. Microsoft builds every ExpressRoute circuit from two connections and, for maximum resiliency, recommends placing two circuits in two different peering locations. Even when the cloud side is prepared this way, a single router in your rack remains a single point of failure. The redundancy you choose also shows up in the price.

What sets the price

Each connection is priced on different factors.

  • Cross Connect: charged as a monthly fee plus a setup fee per connection. The price differs by data center, and within one data center it is the same for every cable type.
  • Metro Connect: no fixed price list. Quoted individually on the two data centers, bandwidth and redundancy.
  • Cloud Connect: no fixed price list. Quoted individually on the cloud provider, data center, bandwidth and redundancy.

For Metro Connect and Cloud Connect, the values you set yourself are bandwidth and redundancy, and you need your normal and peak transfer volumes to set either one.

The order for choosing a connection

For a direct physical link between two devices in the same data center, consider Cross Connect. For a connection beyond the data center, work through these questions in order.

  1. Is a VPN over the internet enough?Check normal and peak traffic and how much latency variation the workload can tolerate. If the existing VPN meets those requirements, there is no need to add a dedicated circuit.
  2. Which endpoints need a dedicated connection?Consider Metro Connect between data centers in the Seoul metropolitan area, or Cloud Connect between your rack and a public cloud.
  3. Tolerable downtimeChoose a single or a dual path by how long the service can hold out, and bring the equipment in your rack up to the same level.
  4. Layer and operating scopeDecide whether you will run the routing yourself, and if you need encryption, add a VPN on top of the dedicated circuit.

Where there are two or more stretches, chain the connections. Metro Connect is also used for hybrid cloud integration, so if the route to the cloud passes through another data center, consider it together with Cloud Connect.

Replication lag at a particular time of day does not, on its own, tell you what caused it. First check whether latency along the network path also rose when replication slowed.

View Infrastructure products

More insights

All insights

Ready to get started?

Tell us about your workload and we will help you choose the right products and plans.

  • Talk to our team

    Questions about anything you read here? Our team will help.

    Contact Sales
  • Insights

    Where PacketStream is heading, what we learn running the network, and the latest product news and notices.

    All insights

Privacy settings

Manage what you allow

Necessary cookies and similar technologies (such as local storage) are always active. You can choose the rest below and change your choice at any time from Cookie settings at the bottom of the site. See our Privacy Policy.

Necessary

Always active

Required for the site to work and stay secure. They cannot be turned off.

Examples: Theme setting, consent record, contact form security check (Cloudflare Turnstile)

Analytics and performance

Optional

Shows how visitors arrive and use our pages so we can improve the site.

Examples: Google Analytics, Microsoft Clarity

Marketing

Optional

Measures how our ads perform and helps show more relevant ads on other sites.

Examples: Google Ads