Public, Private or Hybrid Cloud: Which Workload Fits Where?

A UAE manufacturer wants to update its production systems and give customers a better order portal. The portal gets bursts of traffic when orders open. On the factory floor, operators need to record work even if the external network goes down. Finance needs a reliable copy of shipment and billing records. A single infrastructure choice may suit one of those needs and make another harder to meet.

Start with the smallest unit of work that can operate coherently. Decide which functions must stay close together, what can continue during a disruption and who can support the arrangement on a normal shift. A public service may fit the order portal, a local environment may fit plant operations and a link between them may carry shipment updates. The link deserves the same design and testing as either environment.

Define what the supplier is offering

NIST describes public cloud as infrastructure offered for broad public use and private cloud as cloud infrastructure reserved for one organization. A private cloud may be run by the company or a third party, on or off the company's premises. NIST uses hybrid cloud for distinct cloud infrastructures connected in a way that enables movement of data or applications. In business conversations, hybrid also often describes a mix of cloud and conventional local systems. Ask what the proposal actually connects and which functions it can move. NIST definition of cloud computing

A virtual network inside a public cloud may be private to the customer in the networking sense. Ask whether the provider is offering isolated network access, dedicated hardware or a cloud reserved for the organization. Each describes a different property. Similarly, a server in the plant does not automatically provide cloud features such as pooled resources or provisioning on demand. Get the bidder to draw the proposed arrangement and show what the team can configure itself.

Use the UAE Cyber Security Council's National Cloud Security Policy as a reference when asking about governance, access, resilience and portability. The official summary describes areas for cloud consumers and providers. Have the accountable team determine the provisions and other requirements relevant to the organization and workload, then convert those into evidence requests for the proposed design. UAE National Cloud Security Policy

Place workloads by their operating constraints

List the functions the business performs and the information each needs. The manufacturer might have a public product catalogue, a portal that accepts orders, equipment control on the factory floor and an accounting system. Map which dependencies need an immediate response and which can exchange data later. The equipment control loop and a daily billing export have different tolerances for a slow or unavailable connection.

Consider where the authoritative record lives. If the public portal receives a request, it can show that the request was accepted for processing. A separate confirmation may depend on inventory and production capacity. The portal should distinguish those states if the plant is offline. Giving both environments the authority to promise the same remaining stock creates a reconciliation problem when they reconnect.

Some requirements will rule out placements early. If an operator needs to complete a task while disconnected, test the task without the external link. If a contract limits where specific data can be processed, check each environment and its backup path. If a system depends on a peripheral or license server at the site, ask whether that dependence can change. Separate a hard constraint from a preference for the infrastructure the team already knows.

Count the work each design leaves with your team

For a public-cloud option, identify which operations the provider performs and which stay with the customer. A managed database can reduce some maintenance work while the business still has to configure access, verify backups and check that applications recover. Confirm the proposed service level and which activities need a separate support plan.

For a private deployment, price the hardware or hosting, spare capacity, licensing, patching, monitoring and staff coverage. Ask who supplies replacement equipment and how quickly it can be installed. Dedicated resources can make capacity easier to account for. The business may also pay for that capacity through quiet periods and need lead time to add more. A third party can operate the infrastructure if its contract covers the necessary duties and the company can verify the result.

For a hybrid design, budget for both sides and the connection. Include duplicate monitoring, identity integration, data movement and people who can diagnose a fault that crosses environments. Price recovery independently of normal operation. The system may meet its uptime objective only if the buyer funds a second environment, additional connectivity or a manual procedure for the affected work.

Imagine the manufacturer budgets AED 24,000 a month for its hosted components and AED 16,000 for the local environment, with AED 8,000 for shared connectivity and integration support. That is AED 48,000 a month before one-time work and other exclusions. If the company later removes AED 5,000 in hosting costs but has to add AED 7,000 in local coverage and support, monthly spending rises by AED 2,000. The numbers are illustrative. Use the same scope and service targets for each option, and separate cash spending from staff effort already covered by payroll.

Test the connection as part of the workload

A du campaign reel shows a network demonstration in the UAE. For the manufacturer, a relevant test starts with its own devices, circuits and traffic. Measure how long it takes to confirm an order, submit a batch or retrieve an inventory figure from the locations where staff work. Record latency and failures at different times, then repeat the exercise on the backup link.

Ask which functions depend on a remote control service. A platform can run equipment at the site while still relying on a connection for monitoring, management or updates. AWS documents a service link that connects Outposts to its home region for management and traffic. Use that example to ask suppliers to show which parts of their specific proposal work during a link interruption and which require the remote service. Product behavior and availability need separate confirmation. AWS Outposts service link documentation

Unplug the link in a controlled test. Check whether workers can sign in, whether the local service can read its configuration and whether new events are retained safely. After reconnecting, verify that delayed events are applied once, in the appropriate order. Include a conflict, such as a customer changing an order while the factory system records it as started. Assign a person to resolve the conflict. Give them access to the timestamps and source records from both sides so they can establish which event happened first.

Set an alert for the age of unsent work as well as whether the connection is up. A live link may still carry events too slowly to clear the queue. Suppose a plant sends 120 updates an hour during an outage lasting two hours. It has 240 delayed updates when the link returns. If the receiver can handle 180 per hour while new updates arrive at 120 per hour, the backlog takes four more hours to clear at those rates. Decide whether customer-facing stock or order information must be marked as delayed while reconciliation catches up.

Measure what crosses the link and why. The portal might need available-to-promise quantities and order status, while detailed machine readings can stay with the production team. Ask both teams which fields they use and how stale those fields can become before a decision is unsafe. A smaller exchange can reduce bandwidth and simplify support, provided downstream employees can still verify the record. Document the exchange format and what happens when one system updates it before the other. Include that mismatch in the same acceptance test as disconnection.

Keep the control and recovery responsibilities visible

Make an operating map for identity, logs, keys and backup data. Identify who can grant access in each environment and who investigates when something crosses the boundary. A support engineer may need enough access to trace an event without being able to edit factory or finance records. Check that terminating a user's access reaches both environments and that service accounts have only the operations they need.

Rehearse an outage on each side. If the plant is unavailable, determine whether the portal can keep accepting requests, and decide what it should tell customers about expected confirmation. If the public service is unavailable, check which local work must continue and how the business informs staff. Recover a representative record and compare it with the source. Restoration is complete when the relevant employee can do the job and reconcile any work created during the interruption.

Name the person who owns an incident spanning both sides. A connectivity vendor might confirm that the circuit is reachable while the platform operator reports that its services are running. Give the incident owner the transaction references and timing needed to identify the missing step. Test escalation with the same support tiers and hours the company intends to buy.

Approve the placement one workload at a time

For the hypothetical manufacturer, the portal could sit in public cloud if it handles demand and satisfies its access and data constraints. Plant functions that must run during external disruption could remain in a suitable local environment. The connection could exchange confirmed events with defined rules for delays and conflicts. Finance can then judge the resulting records as a separate workload and check whether it needs a different recovery target.

Compare the viable designs against the same workload, fault scenarios and operating budget. Ask the owners to verify an export or rebuild procedure before the company depends on a provider-specific configuration. Record which team performs each maintenance task, what the company pays for capacity and what happens when a connection or environment fails. If the hybrid option needs a support function the organization cannot fund, change the boundary or simplify the design. Revisit placement when demand, requirements or the dependency map changes. The decision should tell the next owner what must run where, why it belongs there and how to prove the arrangement still works.