Workload predictability: The key to infrastructure decision-making

In today's dynamic world of technology, enterprises have an unprecedented array of powerful options at their disposal. Infrastructure options like “sovereign cloud”' and “private 5G” offer new possibilities and can create a perception of isolation, while a spectrum of infrastructure models- from public and private cloud to on-premise solutions-provides remarkable flexibility. Navigating these choices to find the optimal path for your specific needs is a strategic opportunity in building an effective technology strategy.

This post provides some perspective on how we can navigate this complex infrastructure landscape. It starts with understanding one simple, fundamental distinction: the difference between dedicated and shared Infrastructure. Understanding this single concept is the key to unlocking all other infrastructure choices and to making the most effective, economical decision.

Defining the main infrastructure models: Dedicated, shared and dedicated-like

The distinction is straightforward:

• Dedicated infrastructure: An asset for your exclusive use. It offers unparalleled privacy, security, reliability and guaranteed performance but comes with an upfront investment (CapEx) and ongoing operational costs/XaaS(As a Service). Think of it like owning a private jet.
• Shared infrastructure: A resource used by multiple parties, e.g. public cloud platforms or telecommunications network services. It offers agility, scalability and cost-effectiveness with no upfront investment. Think of it like flying on a commercial airline.

There are some nuances to the dedicated infrastructure (i.e."dedicated-like"/logically isolated, single tenant experience). An example of it is 5G network slicing. For example, a Telco/MNO can provide a private "VIP lane" on its public network to guarantee performance for an enterprise. However, while it feels dedicated, it is still a service running on a single provider's shared infrastructure. It does not provide carrier diversity if that single carrier's network fails, the private lane is also affected. Regardless of the underlying access network technology (4G/5G the transport network is shared most of the time). Network slicing creates a logically isolated, single-tenant experience running seamlessly over a multi-tenant transport network.

Therefore, for mission-critical operations, true resilience comes from a multi-layered strategy. Since any end-to-end connection from a device to an application server will inevitably use some shared components, you must build robust resilience by leveraging multiple network technologies (such as cellular, fiber, satellite etc.,) from multiple, independent carriers to enable network independent access to emergency services. True resilience requires more than just redundant hardware; it requires redundant connectivity pathways. This is where the PACE framework, originated for military communications planning to ensure connectivity in the most challenging environments, becomes invaluable for modern infrastructure planning. PACE stands for Primary, Alternate, Contingency, and Emergency. It provides a structured approach for architecting resilient connectivity by intentionally layering independent network technologies and carriers.

Making decision based on the workload’s predictability:

One way to choose between these infrastructure types is to analyze the nature of the workload the infrastructure will support. Is it dynamic and unpredictable, or is it static and predictable?

The economics of uncertainty: Using shared infrastructure for unpredictable needs

An unpredictable workload is one that is used intermittently, on an ad-hoc basis, or sits idle for most of its lifecycle. Building dedicated infrastructure for a workload that is 99% idle isn't economical.

Consider the challenge of coordinating multiple public safety agencies during a major incident. Instead of each agency buying and maintaining expensive, dedicated gateway hardware that sits idle most of the time, they can instantly spin up a communication gateway in the cloud.

Critical Connect is a cloud-based solution that connects different radio systems (e.g. Tetra, P25). This is a perfect use case for shared infrastructure:

• No upfront investment: Eliminates the need for heavy CapEx on hardware.
• Pay-as-you-go: Pay only for the hours or days the communication gateway is needed.
• Instant scalability: Support ten users or ten thousand on demand.

The agility and cost-effectiveness of this model are undeniable for temporary needs (dynamic demand) and disaster recovery scenarios.

Predictable workloads: Dedicated vs shared infrastructure

A static workload is constant, predictable and essential for an organization's daily operations. Think of public-safety agency dispatch software or a hospital's electronic health record system. These "always-on" systems are business-critical workloads. The capacity requirements are also quite predictable. In the public safety scenario, the exception could be unplanned incidents/large events.

For predictable, high-utilization workloads, dedicated infrastructure seems to be the optimal choice. The primary concerns shift from on-demand flexibility to achieving the highest form of security, reliability, performance and control.

A dedicated model offers:

• Maximum security, reliability and privacy: Physically isolating your hardware eliminates risks associated with sharing resources.
• Guaranteed performance: The "noisy neighbor" effect present in some shared environments is completely removed.
• Total control: Your organization has full control over the hardware and software.

So when to use shared infrastructure for predictable workloads? There are other factors influencing the use of shared infrastructure for predictable workloads, e.g. integration cost and complexity.

Navigating the terminology: Other common infrastructure models

Now that the main concept is established, let’s see the other common ways to classify infrastructure. The three other ways to classify infrastructure are by its location (where it lives), its service model (how it's delivered) and increasingly by its architectural philosophy.

• Location (where it lives): A dedicated system can be on-premise in your own data center, while a shared system is typically in an external location (e.g. cloud/service provider’s data center). There are some nuances to the hyperscaler edge propositions (e.g. Azure Edge, AWS Outpost, AWS Wavelength etc.) though installed in a customer’s data center they fall under the category of dedicated-like infrastructure.

Depending on the capacity/scale, it can be classified as:

• Core/centralized infrastructure: Traditional or hyper scale data centers that provide the bulk of cloud services.
• Edge infrastructure: Compute and storage resources are placed closer to the source of data generation or the end-user (e.g. in a factory, a retail store).
• Service model (what you manage vs. what the provider manages): A cloud service is a shared Infrastructure as a Service (IaaS/PaaS/SaaS solution. Dedicated private cloud service managed by a service provider.
• Architecture: These classifications reflect modern DevOps practices and architectural trends.
• Mutable infrastructure: The traditional approach. Servers are continuously updated, patched and modified in place. They are treated like "pets" that are cared for over a long time.
• Immutable infrastructure: The modern cloud-native approach. Servers are never modified after deployment. If an update is needed, the existing server is terminated and a new one with the latest configuration is deployed to replace it. They are treated like "cattle," not individuals.

A hybrid strategy

From an enterprise (e.g. law enforcement agency) customer perspective, the choice between shared and dedicated infrastructure is not about the technology itself, but about matching the right hosting model to the right workload. The most effective infrastructure strategy is rarely a rigid choice of one model over the other, but a hybrid approach tailored to your workloads. An organization can run its predictable workloads on a secure, dedicated infrastructure while leveraging shared, on-demand services for unpredictable events that require external collaboration.