Choosing a Colocation Home

Choosing where to physically house your servers is one of the most consequential decisions you will make as an infrastructure operator. It determines your latency floor, your uptime ceiling, your operational overhead, and your monthly burn rate. Get it wrong, and you are locked into a contract with a facility that cannot support your growth. Get it right, and your infrastructure runs quietly in the background while you focus on building.

This is what we learned while selecting and operating datacenters for Honeycluster.

Cost

Colocation pricing is not as simple as a monthly rack fee. The real cost is a composite of several line items that vary wildly between providers:

•
Rack space — Full cabinet, half cabinet, or per-U pricing. Density matters more than raw space.
•
Power — Metered (pay for what you draw) or committed (pay for a fixed allocation whether you use it or not). Overages are expensive.
•
Bandwidth — Committed data transfer, burstable capacity, and whether the provider charges for inbound, outbound, or both.
•
Cross-connects — Physical connections to other networks or providers within the facility. These add up quickly.
•
Remote hands — Charges for datacenter staff to physically interact with your equipment. Some providers include a monthly allotment; others bill per incident.
•
IP addresses — Static IPv4 addresses are a finite resource. Expect to pay per IP, and plan for how many you actually need.

Watch for hidden costs

Setup fees, contract termination penalties, and power overage charges can significantly inflate your effective monthly cost. Always model the worst-case scenario, not just the quoted rate.

Location

Geographic placement affects three things simultaneously: latency to your users, latency to the network you are serving, and your ability to physically access the facility when something goes wrong.

For XRPL infrastructure, proximity to major internet exchange points and peering partners matters more than proximity to end users. Your load balancer handles geographic distribution to users — your nodes need low-latency connections to the peer-to-peer network.

Practical considerations:

•
How far is the facility from your team? If a drive fails at 2 AM, can someone be on-site within an hour?
•
Is the facility in a region prone to natural disasters? Flood zones, hurricane paths, and earthquake-prone areas add risk that no amount of redundancy inside the building can fully mitigate.
•
What is the local power grid reliability? Some regions have significantly more stable utility power than others.

Power redundancy and failover

Power is the single most critical dependency for any physical infrastructure. When the power goes out, everything goes out — and the recovery time depends entirely on how the facility is designed.

A properly redundant power path looks like this:

•
Utility power from the grid (ideally from two independent feeds)
•
UPS (Uninterruptible Power Supply) that absorbs the gap between utility loss and generator startup — typically 10-30 seconds
•
Diesel generators that can sustain the facility for hours or days depending on fuel capacity
•
Automatic Transfer Switch (ATS) that handles the cutover without manual intervention

Know your rack's power budget

Our rack is limited to a 20-amp circuit, which we tripped several times while running our ScyllaDB nodes under heavy write loads. The bottleneck was amperage, not watts — at 110V, a 20-amp circuit only supports roughly 2.2 kW (watts = volts × amps). Exceeding your circuit's amp rating will trip breakers and take your entire rack offline. Understand your peak current draw, not just your average, and leave headroom for growth.

Key questions about power

###

•
What is the N+1 redundancy on generators? If one generator fails, is there a backup?
•
What is the refueling interval? A 24-hour fuel supply is standard. Less than that is a risk during extended outages.
•
Are the CRAC units (air conditioning) on generator power? If the generators run but cooling does not, thermal shutdown will take your servers offline within minutes.
•
How often are generators and UPS systems tested? Monthly load testing is the minimum. Ask for maintenance records.

Access to the facility

Physical access policies vary dramatically between providers. Some offer 24/7 badge access with no notice required. Others require advance scheduling, escort accompaniment, and limited hours.

For production infrastructure, you need:

•
24/7 unescorted access — Emergencies do not happen during business hours
•
Reasonable security — Biometric or badge access, mantrap entries, security cameras
•
Loading dock access — For receiving and shipping hardware
•
Comfortable working conditions — Adequate lighting, workbenches, and network access in the cage area

Remote hands as a backup

Even with 24/7 access, having reliable remote hands support is essential. When your team is unavailable or the issue is minor — a cable reseat, a power cycle, a visual inspection — remote hands saves a trip. Confirm the provider's response time SLA and whether they charge per incident or include a monthly allotment.

Bandwidth and connectivity

Not all datacenter bandwidth is created equal. The quality of a facility's network connectivity determines your baseline performance and your ceiling during traffic spikes.

What to evaluate

###

Factor	Why It Matters
Carrier diversity	Multiple upstream providers prevent a single carrier outage from isolating your infrastructure
Internet exchange presence	Direct peering at an IX reduces latency and transit costs
Static IP availability	You need predictable, routable addresses for your services
Number of drops per rack	Each physical network connection to your rack is a "drop" — plan for current and future needs
Burstable capacity	Can you handle traffic spikes without pre-purchasing committed bandwidth?
DDoS mitigation	Does the facility or upstream provider offer volumetric attack protection?

Managed vs. unmanaged connectivity

###

Some facilities provide a managed network — you plug into their switches and they handle routing, firewalling, and IP allocation. Others give you a raw cross-connect and you bring your own network equipment.

For infrastructure like Honeycluster, we prefer unmanaged connectivity. We control our own routing, our own firewall rules, and our own IP space. This adds operational complexity but gives us full control over traffic flow, peering decisions, and security policy.

Questions to ask before signing

Before committing to a datacenter, ask these questions directly. The answers will tell you whether the facility is a long-term partner or a liability:

•
Are you a reseller or do you own the facility? — Resellers add a margin and a layer of indirection. When something breaks, you want to be talking to the people who control the building.
•
Do you manage your own carrier contracts? — Facilities that own their network relationships can resolve connectivity issues faster than those who depend on third parties.
•
What is your routing infrastructure? — Understand the network topology. How many upstream providers? Is there BGP redundancy? What happens if a transit provider goes down?
•
Are the CRAC units on generator power? — If the answer is no, walk away. Servers without cooling will thermal-throttle and shut down within minutes of a power event.
•
What is your generator refueling interval? — Anything less than 24 hours of on-site fuel is a risk during extended utility outages.
•
How often are generators and UPS systems maintained? — Ask for maintenance logs. Monthly testing under load is the standard.
•
What happens when the power goes out? — This is deliberately open-ended. You want to hear a detailed, practiced answer — not hesitation.
•
Is this a long-term contract or month-to-month? — Long-term contracts usually come with better pricing, but lock you in. Month-to-month gives flexibility but costs more. Know what you are committing to.

What we learned

Choosing Honeycluster's datacenter facilities was not a single decision — it was a series of tradeoffs evaluated against our specific requirements: high power density for database nodes, low-latency connectivity to XRPL peers, 24/7 physical access, and room to grow.

We explored three different colocation providers in the North Dallas area before making a final decision. Each had strengths, and each had dealbreakers:

•
Facility A had existing rack infrastructure that would not meet our power and density requirements. They offered to fund the cost of a new rack installation in their facility to accommodate us, which was a strong gesture — but the ongoing monthly costs were significantly higher than the other options. When you factor in the long-term burn rate, an upfront concession on installation does not offset years of elevated recurring charges.
•
Facility B was in a very secure location — gated access, staffed front desk, proper physical security throughout. The facility itself was well-maintained. The problem was operational: they did not offer 24/7 access, and our racks would be housed in a larger shared room where our equipment was not the priority. When you are sharing floor space with larger tenants, your tickets go to the bottom of the queue.
•
Facility C offered a half rack with a single ethernet drop, five static IPs, a 20-amp circuit per rack, and unmanaged bandwidth up to 1 Gbps. It is a smaller datacenter located in the core of an office building — not the kind of place that looks impressive on a tour, but the kind of place where the operators know every piece of equipment in the room. The team was hands-on, offered practical advice on our setup from day one, and gave us 24/7 unescorted badge access.

We went with Facility C. The deciding factors were not the flashiest specs — they were the human ones. The operators gave us honest answers, practical guidance, and the flexibility to grow at our own pace. In a smaller facility, you are not a ticket number. You are a tenant they actually know.

No facility is perfect. The goal is to find one where the compromises are acceptable and the operator is transparent about limitations. A datacenter that honestly tells you their circuit caps at 20 amps per rack is more trustworthy than one that promises unlimited power and trips your breakers three months in.

The bottom line

Your infrastructure is only as reliable as the building it lives in. Spend the time upfront to evaluate facilities thoroughly — tour in person, ask hard questions, and talk to existing tenants. The cost of choosing wrong is measured in downtime, emergency migrations, and lost trust.

Components of the Hive

The invisible layer

A successful launch week - by the numbers