What is colocation?
Colocation is a data center service in which a customer rents space, power, cooling and network access in a facility run by a provider, and installs servers, storage and network equipment that the customer owns. The provider operates the building and its mechanical, electrical and physical security systems. The customer operates the IT: hardware configuration, operating systems, applications and most maintenance. Colocation lets organizations use professionally run facilities without building their own data center.
Retail vs wholesale colocation
Retail colocation is sold by the rack, partial rack or cage, with power provided as fixed circuits or as metered usage. Contracts tend to be shorter and include shared infrastructure. Wholesale colocation leases a larger dedicated area, such as a data hall or suite, with committed power capacity in the hundreds of kilowatts or more. Wholesale customers usually take on more of the fit-out and pay mainly for reserved power. The boundary between the two varies by provider and market.
Colocation vs cloud and managed hosting
Colocation sits between owning a data center and renting compute. The difference is who owns and operates each layer, from the building and power systems up through servers, operating systems and applications. Many organizations use more than one model at once, for example colocation for steady core systems and public cloud for variable or seasonal workloads.
- Own data center: you own the building and the IT.
- Colocation: the provider runs the facility; you own and run the IT.
- Managed hosting: the provider supplies and maintains the servers; you run the applications.
- Public cloud: the provider runs everything up to a service interface; you pay for capacity as you use it.
Worked example: planning rack power
Power is usually the binding constraint. Take a rack with redundant A and B feeds, each a single-phase 32 A circuit at 230 V, or about 7.4 kW per feed. Many providers limit continuous draw to around 80 percent of the breaker rating, which gives roughly 5.9 kW. Because either feed must carry the whole rack if the other fails, the combined A and B load should stay below 5.9 kW. Adding the two feeds together to 11.8 kW is a common planning mistake. Fourteen servers at 400 W measured draw use 5.6 kW and fit; sixteen would not.
What to check in a provider
Provider selection affects cost and risk for years, and moving racks out of a facility later is slow and disruptive. Compare more than the price per rack: visit the site, ask for recent maintenance and incident records, and confirm the contract terms and service credits behind each item below.
- Facility resilience, such as an Uptime Institute Tier certification or an EN 50600 availability class.
- Maximum power density per rack and how power is billed.
- Cooling design, including liquid cooling support for high-density GPU racks.
- Carrier choice, cross-connect pricing and access to internet exchanges.
- Remote hands scope, response times and hourly charges.
- Security certifications such as ISO/IEC 27001 and the access process for your staff.
Operating hardware you cannot see
Colocated equipment may sit in another city or country. Each site visit costs travel time, and remote hands staff follow instructions without knowing your environment. Operators therefore depend on out-of-band access through server BMCs for power cycling and console access, on asset records that state the rack, unit and serial number of each device, and on hardware alerts that arrive before a component fails. Clear runbooks for remote hands, with photos and labels, reduce mistakes such as pulling the wrong disk.
Label every device on the front and back with the same identifier used in your asset records before shipping it to the facility.
How CloudSino approaches colocation
CloudSino supports colocation operations through DCOS and iDCOS. DCOS connects to server BMC management ports, including iLO, iDRAC, iBMC and IMM, over IPMI and Redfish, and provides remote power on, power off, restart, KVM over IP console and batch control for colocation sites and unattended server rooms. It records hardware changes, component replacements and device racking and unracking. iDCOS supports operations across headquarters server rooms, branch rooms, colocation sites, disaster recovery centers and cloud platforms, with ITSM workflows and a CMDB.
Frequently asked questions
Who is responsible for hardware failures in colocation?
The customer, in most contracts. The provider is responsible for the facility, power and cooling up to the agreed demarcation point, usually the rack PDU or power socket. Replacing a failed disk or server is the customer's task, done by its own staff, a maintenance partner or the provider's remote hands under instruction.
What are remote hands?
Remote hands is an on-site service in which provider staff perform simple physical tasks for a customer, such as power cycling a device, reseating a cable, swapping a labeled part or reading a front panel display. It is usually billed per task or by time, so out-of-band access that avoids these calls saves both.
Is colocation cheaper than public cloud?
It depends on the workload. Steady, predictable workloads with high utilization often cost less on owned hardware in colocation, while variable or short-lived workloads can suit cloud pricing. A fair comparison includes hardware depreciation, staff time, network costs, licensing and the cost of spare capacity.