Muokkaa

Jaa

Operational Excellence tradeoffs

  • Artikkeli

Operational Excellence provides workload quality through the implementation of clear team standards, understood responsibility and accountability, attention to customer outcomes, and team cohesion. The implementation of these goals is rooted in DevOps, which recommends minimizing process variance, reducing human error, and ultimately increasing the return of value for the workload. That value isn't just measured against the functional requirements served by the components of the workload. It's also measured by the value that the team delivers in striving for improvement.

During the design phase of a workload and over its lifecycle, as continuous improvement steps are taken, it's important to consider how decisions based on the Operational Excellence design principles and the recommendations in the Design review checklist for Operational Excellence might influence the goals and optimizations of other pillars. Certain decisions might benefit some pillars but constitute tradeoffs for others. This article describes example tradeoffs that a workload team might encounter when designing workload architecture and operations.

Operational Excellence tradeoffs with Reliability

Tradeoff: Increased complexity. Reliability prioritizes simplicity, because simple design minimizes misconfiguration and reduces unexpected interactions.

  • Safe deployment strategies often require some amount of forward and backward compatibility between application logic and data in the workload. This added complexity increases the testing burden and can lead to complexities or integrity issues with the workload's data.

  • Highly layered, modularized, or parameterized infrastructure as code can increase the chance of accidental misconfiguration because of the complexity of the interaction between the code components.

  • Cloud design patterns that benefit operations sometimes necessitate the introduction of additional components, for example, the use of an external configuration store or the coordination of sidecar deployments in a containerized application platform. The additional components and added layers of indirection increase the points of interaction in the system, increasing the surface area for malfunction or misconfiguration.

  • Workload components that are designed to independently evolve to support agile development and hosting introduce dependencies on service discovery as a layer of indirection. Service discovery might lack responsiveness to change, and malfunction can be hard to diagnose.

Tradeoff: Increased potentially destabilizing activities. The Reliability pillar encourages the avoidance of activities or design choices that can destabilize a system and lead to disruptions, outages, or malfunctions.

  • Deploying small, incremental changes is a technique for mitigating risk, but those small changes are also expected to be delivered to production more frequently. Deployments can destabilize a system, so as the rate of deployment increases, so does this risk.

  • A culture that measures itself with velocity metrics like deployments per week and uses automation that can facilitate introducing changes at a faster pace is also likely to perform more deployments in a shorter period.

  • Increasing density to simplify operations by reducing the number of control and observability surfaces can also lead to an increased availability risk because malfunction or misconfiguration increases the impact radius of a destabilizing event.

Operational Excellence tradeoffs with Security

Tradeoff: Increased surface area. The Security pillar recommends a reduced workload surface area in terms of components and exposure to operations. This reduction minimizes attack vectors and produces a smaller scope for security control and testing.

  • Components that surround the workload and support its operations, like automation or a custom control plane, must also be in scope for regular security hardening and testing.

  • Routine, ad hoc, and emergency operations increase the points of contact with the workload. A zero trust approach requires that these processes are considered attack vectors and must be included in the security controls and validation for the workload.

  • The observability platform of the system collects logs and metrics about the workload, which can be a valuable source of information disclosure. Therefore, the workload's security needs to extend to protect data sinks from internal and external threats.

  • Build agents, externalized configuration and feature toggle stores, and side-by-side deployment approaches all increase the application surface area that requires security.

  • A higher deployment frequency caused by small, incremental changes or by "get current, stay current" efforts results in more security testing in the software development lifecycle.

Tradeoff: Increased desire for transparency. A secure workload is based on designs that protect the confidentiality of data that flows through the components of the system.

Observability platforms ingest data of all types to gain insights into a workload's health and behavior. As teams try to attain higher fidelity in observability data, there's an increased risk that data classification controls, like data masking, of the source systems don't extend to the logs and log sinks of the observability platform.

Tradeoff: Reduced segmentation. A key security approach for isolating access and function is to design a strong segmentation strategy. This design is implemented through resource isolation and identity controls.

  • Co-locating disparate application components in shared compute, network, and data resources to make management easier reverses segmentation or makes role-based segmentation harder to achieve. Co-located components might also need to share a workload identity, which can lead to over-assignment of permissions or a lack of traceability.

  • Collecting all logs from across the system in a unified log sink can make querying and building alerts easier. However, doing so can also make it harder or impossible to provide row-based security in order to treat sensitive data with the required audit controls.

  • Simplifying the management of attribute-based or role-based security by reducing the granularity of roles and their assignments can lead to inappropriately broad permissions.

Operational Excellence tradeoffs with Cost Optimization

The Operational Excellence pillar never recommends activities that reduce productivity or jeopardize a workload's return on investment. Recommendations that seem to shift focus from delivery activities take into account long-term best interests for the workload and team. If your workload is nearing its sunset date, it probably doesn't make sense to invest highly in recommendations that trigger these tradeoffs.

Tradeoff: Increased resource spending. A major cost driver for a workload is the cost of its resources. Deploying fewer resources, right-sizing resources, and reducing consumption generally helps keep costs low.

  • Implementing safe deployment practices, even if the changes are relatively small, can lead to an increase in the number of resources that are concurrently deployed. These patterns require the deployment of multiple concurrent instances of the application or infrastructure component so that traffic can be shifted in a controlled way. This increase is more pronounced in a workload that uses an immutable infrastructure approach.

  • The team might need to introduce additional workload components in order to implement operationally aligned cloud design patterns or workload automation. For example, to support deployment agility, they might add a gateway routing component. To support better configuration management, they might add an external configuration store. To support tenant lifecycle events, they might build a control plane. These resources also influence the costs of preproduction environments.

  • Increasing the number of preproduction environments to improve the development and testing experience through isolation also increases the number of resources. These resources, which aren't used to deliver supply against production demand, increase the cost of the solution.

  • Increasing the parity of preproduction environments with the production environment, in terms of resource count, SKUs, and data volumes, improves the quality assurance process. The cost increases as parity increases.

  • Although telemetry data isn't directly a resource, to enable the effectiveness of observability platforms, this data needs to be persisted. Most operational data stores have pricing that's based on a combination of ingestion rates and volume. Generally, as the amount of low-latency, high-diversity telemetry increases, costs also increase. For multi-region deployments, these operational data sinks are expected to be deployed per region, so any per-resource costs become a factor.

Tradeoff: Decreased focus on delivery activities. Workload team members deliver increased workload value by efficiently performing tasks that are aligned to their capabilities.

  • Workload teams that spend time creating and refining a healthy and responsible support structure and incident response are providing a valuable service to the workload's users. As the support effort increases (for example, formal on-call rotations), usually because of a change in business criticality, the costs of these activities increase. This cost increase can be the result of an increase in staff or can be incurred indirectly in the form of attention that's shifted from delivery activities to supporting functions.

  • Training is a critical part of a workload team's personal continuous improvement process. This training can be formal or self-directed during personal enrichment time. As the amount of training time increases, the amount of time available for direct development of the workload decreases. Investment in training is diminished when the training isn't role-based or specifically relevant to the workload or its future.

  • Standardized routine operational tasks for protecting the reliability, security, and performance efficiency of a workload take time to define, refine, and perform. This time isn't directly spent on delivery. Some examples of these tasks are comprehensive change impact analysis, change control processes, thorough testing, and increased patch management. As the frequency, comprehensiveness, or operational burden of these tasks increases, the time invested also increases.

Tradeoff: Increased tooling demands and diversity. The Cost Optimization pillar recommends the reduction of tooling sprawl, consolidation of vendors, and a right-sized approach to all tooling purchases.

A workload team purchases tools and hardware to support activities that are performed during the entire software development lifecycle (SDLC), including planning and design, development and testing, and monitoring. The marketplace for tooling in this space is growing. Tools are offered at various price points that usually correspond to the tools' features and capabilities. Except for free offerings, these tools incur initial licensing costs, which might be per-user, per-device, or site-wide. They often also require ongoing maintenance contracts. New vendor relationships might need to be established. Here are some examples of expected tooling or hardware spending that's associated with the principles of operational excellence:

  • Requirements and backlog management
  • Architecture design tools
  • UI/UX design tools
  • Code and asset hosting
  • Code and low-code development environments
  • Automation tools
  • Development and quality assurance workstations
  • Development and deployment pipelines
  • Test execution and tracking
  • Observability tools

Operational Excellence tradeoffs with Performance Efficiency

Tradeoff: Increased resource utilization. The Performance Efficiency pillar recommends the allocation of as much of the available compute and network as possible to the requirements of the workload.

  • A workload's observability framework requires that the components in the architecture allocate time and resources to create, collect, and stream logs and metrics. These data points help ensure that effective alerting and monitoring is possible for reliability, security, and performance. As the level of instrumentation increases, the strain on system resources might also increase.

  • Some deployment models, like blue/green deployment, which a workload might use for safe deployment, might introduce side-by-side deployments on the production application platform. These deployments require preemptive scaling to provide enough supply to meet future demand, or leave a mostly dormant deployment in place for a period of time to support rollback.

Tradeoff: Increased latency. To create performant workloads, teams look for ways to reduce the time and resources that workloads consume to perform their tasks.

  • Many deployment models require the use of gateway routing access patterns, which can introduce latency. This latency draws against the performance target budget for the related flows.

  • Some cloud design patterns that support "independent change over time" approaches to support the ideals of incremental improvement can introduce latency due to the traversal of additional components. This latency can be introduced by gateways, messaging brokers, or anti-corruption layers.

Explore the tradeoffs for the other pillars: