A hybrid cloud migration is a migration whose target comprises two environments: part of the workloads moves to a cloud provider, the rest stays on an infrastructure that the organisation operates itself, and the two continue to exchange data. This split is part of the migration decision.
The workloads that stay on site are those that a named constraint keeps there: data that cannot leave or whose volume makes transfer unreasonable, local equipment to which the application must respond without delay, a licence whose terms change outside the original infrastructure, a dependency on a system that is not moving. The workloads that move are those that none of these constraints holds back and to which the cloud brings variable capacity, a managed service or the end of hardware due for renewal. Each decision to keep a workload is then qualified: lasting, or waiting for a deadline. The second case describes a transitional state, to be given a date.
The link between the two environments becomes a production component
NIST, in its Special Publication 800-145 of September 2011, defines hybrid cloud as the composition of two or more distinct cloud infrastructures (private, community or public) that remain unique entities, bound together by technology that enables data and application portability. Here we use another meaning, the one used on this site: a cloud combined with an infrastructure operated in-house, whether or not the latter is organised as a private cloud.
In both readings, the word that matters is the link. A hybrid migration adds to the information system a component that did not exist: the interconnection between the site and the provider. It carries production flows, with a throughput, a transit delay, a transfer cost and outages. Added to this are identities to federate and monitoring to extend to both sides. The architecture models that organise this coexistence are compared on our page on sovereign and hybrid cloud. Adding a second cloud provider is covered in our article on multicloud strategy.
Running the cutover remains as described in our article on IT migration. What is specific to hybrid is the question asked of each workload: on which side of the link will it live?
The criteria for splitting between the site and the cloud
| Criterion | Question to ask for each workload | What argues for staying on site | What argues for moving |
|---|---|---|---|
| Data | Where are these data entitled, and better off, to reside? | A location or qualification requirement that the intended target does not meet; a volume whose initial transfer and then daily exchanges weigh on the link | Data with no particular requirement, or a cloud service that meets the requirement |
| Latency | What does the workload talk to, and within what delay? | A tight exchange with equipment or a system that remains on site | Users and partners already outside the site |
| Licences | What does the software contract say about the execution environment? | Rights tied to hardware or to a site, unfavourable terms outside the original infrastructure | A portable contract, or software replaceable by a managed service |
| Dependencies | Which systems does the workload call, and which call it? | Tight coupling with a system that will not migrate | A group of workloads that can move together |
A single criterion is enough to keep a workload, provided the reason is written down and verified: an assumed constraint may fall away on reading the contract or the applicable text. The survey is carried out during the inventory of the estate, a step described on our page on cloud migration together with the possible strategies for each application.
What data impose, by requirement or by volume
The word “data” covers constraints of different kinds, which do not call for the same response.
One is a requirement: location, the law applicable to the provider, the qualification level of the service. It does not necessarily point to the site. An obligation to host on a SecNumCloud qualified service, ANSSI’s qualification framework for cloud service providers, can be met at a provider, on a service so qualified. Staying on site is then one response among others.
The other relates to volume. A large data warehouse, archives, images produced by equipment move slowly, and bringing them back involves outbound transfers, whose billing on a change of provider or a return on site is regulated by Regulation (EU) 2023/2854, known as the Data Act. The workloads that read these data continuously are better off staying close to them. The reasoning works both ways: once the data store has moved, the processing that exploits it has an interest in following it.
Latency and dependencies draw groups
Latency is judged on specific exchanges. A workshop controller, a piece of medical equipment, a point-of-sale system expect a response within a delay that the round trip to the provider may exceed. The workload that responds to them then stays as close as possible, and the cloud receives what tolerates the delay: consolidation, analysis, archiving.
Between applications, the same reasoning goes through the flow map. Two systems that talk to each other at every transaction, placed on either side of the link, pay for the trip at every call and depend on its availability. A database queried by many applications or a central system that will not migrate hold around them what depends closely on them.
The split is therefore made by groups: the cut runs where exchanges are infrequent, asynchronous or tolerant of delay. Drawing up this map is a matter of IT architecture, which then serves to order the waves.
Licences and hardware, the constraints missing from the technical inventory
A discovery tool lists servers and flows, without reading contracts. A licence contract may tie rights to hardware, to a number of physical processors or to a site, and provide for other counting rules on shared infrastructure. Reading the contract with procurement and legal therefore comes before classifying the workload. It may conclude in favour of keeping it, changing the licence or replacing the software.
Hardware weighs in the same way. An encryption appliance or a system that runs on none of the architectures offered by the provider cannot be moved as is. Recently purchased equipment argues for keeping it until the end of its depreciation, which is a dated reason. Conversely, the end of hardware support or the end of a data centre lease sets a deadline for moving.
Lasting or transitional hybrid cloud migration, two ways of investing
A lasting hybrid and a transitional hybrid look alike on cutover day and diverge afterwards. The first has reasons with no date: a data requirement that only the site meets, local equipment, a system that the organisation has decided to keep. The second has reasons that expire: a running contract, hardware still being depreciated or an application whose replacement is planned.
The distinction changes the way of investing. A lasting hybrid justifies a redundant interconnection, operating practices common to both sides and teams that master both. A transitional hybrid calls for the minimum that will hold until the deadline, and a schedule. Mistaken for lasting, it makes the organisation pay for two platforms without anyone having decided so, while the technical debt left on site may continue to grow.
Proportions matter too. When the survey keeps almost the entire estate on site, opening a cloud environment for the rest adds a platform, skills and a link to operate for a narrow benefit: postponing then deserves consideration. When only a few workloads remain, the question becomes the cost of a data centre maintained for them.
The decision that closes the exercise therefore concerns the site itself. If it remains for workloads whose retention is lasting, it is sized and funded for them. If it remains only for workloads that are waiting, each is given a review date, and the site a closure date.