← Back to Concept Contrast

Concept Contrast Fable

On-premises deployment and cloud native

Conclusion first:One moved the system into the customer's own machine room; the other was designed from the ground up to thrive only in the cloud. Choosing on-premises deployment doesn't mean fighting cloud-native, but running an on-premises system as if it were cloud-native will send both your bill and your failure rate upward.

One moved the system into the customer's own machine room; the other was designed from the ground up to thrive only in the cloud. Choosing on-premises deployment doesn't mean fighting cloud-native, but running an on-premises system as if it were cloud-native will send both your bill and your failure rate upward.

Fable

The machine moved into the yard and the big tree growing in the yard

Qingyun Carriage House originally ran its main depot beside the Grand Post Road. Carriages and horses, wheelwrights, fodder warehouses and dispatch boards were all in one place; when a carriage was free, the manager could flip a board and send it to the next station. On rainy days, sheds in neighboring counties could also be borrowed temporarily, making the whole post road like a network that could be dispatched.

Later a mine said its goods were valuable and its route confidential, and asked the carriage company to move a set of horses, carriages, and a repair shed into the mine yard. The manager thought it was just a change of location, so he moved a copy in following the main shop's layout.

The first few months went smoothly. The mining site felt secure, outsiders could not get in, and accounts were checked within the yard. But peak season came, and there were not enough horses in the yard; they could not casually borrow empty carts from the main depot, and when the wheel repairman got sick, they could not immediately transfer a craftsman from a neighboring county. The yard was safe, but it lacked the flexibility of the great post road.

The shopkeeper first stuffed more horses, more fodder, and more spare parts into the yard. Costs rose quickly, and it was wasteful in idle times. Only then did the mine manager realize that moving into the yard was not shrinking and copying the main shop, but choosing to keep control, isolation, and operational responsibility local.

The main depot is not simply easier either. It relies on unified scheduling, continuous improvement, on-demand expansion, and cross-station reuse; one upgrade can benefit many places, and one failure can affect a wider area. Each approach has its costs.

Later, before signing, the carriage company would first ask: does the customer want in-house control, data that never leaves the premises, and independent maintenance, or rapid expansion, unified updates, and elastic scheduling. One is like moving the machine into the courtyard, the other like growing the business along the main post road; choosing wrong means miscalculating security, cost, and speed entirely.

The mine later accepted the in-house fleet, but the contract was written more clearly: which maintenance is handled by the dealership's people, which spare parts stay local, when support can be requested from headquarters, and whether upgrades require prior approval from the mine. Independence brings autonomy, but also more local responsibility.

Headquarters customers chose another kind of commitment: they were willing to let the carriage company centrally manage horses, carriages, and routes in exchange for faster expansion, lower idle capacity, and continuous updates. From then on, the shopkeeper no longer packaged the two options as a difference in location, but laid out control, flexibility, cost, and responsibility for customers to choose. Later, when the carriage company quoted prices, it no longer wrote only "put it in your yard" or "put it at headquarters." It listed isolation, upgrades, dispatch, backup, and maintenance staffing item by item. Only when customers understood these did they know which operating model they were buying.

Localized deployment

Install and run the software system on the customer's own physical servers, private data center, or designated infrastructure, so data stays within the customer's control. Operational responsibilities (hardware, network, backup, disaster recovery) are borne by the customer or shared by agreement with the service provider.

Cloud native

An application architecture paradigm that assumes from the design stage that the runtime environment is a cloud platform, with core features including containerization, microservices, declarative APIs, elastic scaling, and reliance on cloud-managed services. It runs most efficiently on the cloud, and migrating to a non-cloud environment requires extensive architectural adaptation.

Story Mapping

  • The unified node specifications in the cloud do not exist in the customer's machine room: cloud-native scheduling policies depend on the infrastructure assumptions of the cloud environment.
  • Health check timeout changed from five seconds to twenty seconds: localized deployment requires re-adapting all cloud-native preset parameters.
  • Turn off autoscaling and switch to a fixed number of instances: the elastic infrastructure that cloud-native depends on does not exist in localized environments.
  • Ops cost comparison table: everything the cloud does for you becomes manual ops cost in an on-premises deployment.
  • Transplanting a large tree requires reblending soil and adjusting watering: moving a cloud-native product to localized deployment is not a move; the key is transplanting.

In the project

In the pre-sales stage, split "deployment method" and "operations mode" into two independent evaluation dimensions. Deployment method determines where the data is and who manages the hardware; operations mode determines who performs routine inspections, scaling and failure recovery, and under what SLA it runs. When a cloud-native product is deployed on-premises, attach a "cloud environment alternative checklist" to the contract, listing item by item how each thing the cloud originally did automatically is implemented in the on-premises deployment and who is responsible. This checklist is far more effective than simply labeling it "supports on-premises deployment."

京ICP备2026014272号