DevOps tech: Cloud infrastructure

Many organizations are adopting cloud computing. But there's more to cloud than buying your infrastructure from a cloud provider. The US National Institute of Standards and Technologies (NIST) defines five essential characteristics of cloud computing:

  • On demand self-service: Consumers can provision computing resources as needed, without human interaction from the provider.
  • Broad network access: Capabilities can be accessed through diverse platforms such as mobile phones, tablets, laptops, and workstations.
  • Resource pooling: Provider resources are pooled in a multi-tenant model, with physical and virtual resources dynamically assigned on-demand. The customer may specify location at a higher level of abstraction such as country, state, or data center.
  • Rapid elasticity: Capabilities can be elastically provisioned and released to rapidly scale outward or inward on demand, appearing to be unlimited and able to be appropriated in any quantity at any time.
  • Measured service: Cloud systems automatically control, optimize, and report resource use based on the type of service such as storage, processing, bandwidth, and active user accounts.

The 2019 State of DevOps Report from DevOps Research and Assessment (DORA) found that only 29% of people who claim to have adopted cloud technologies agreed or strongly agreed that they met all five of the characteristics defined above. In both 2018 and 2019, DORA's research found that elite performers were more than 23 times more likely to have met all five essential cloud characteristics than low performers. This may explain why teams and executives who claim to have adopted cloud computing technologies also express frustration at not reaping the expected results.

Because many organizations still use their traditional datacenter practices and processes to manage their cloud infrastructure, they can't achieve the expected benefits, which according to DORA's research include:

  • Improved software delivery performance: faster throughput and higher levels of stability
  • Better service availability
  • Improved cost visibility: in 2019, we found that respondents who met all essential cloud characteristics are 2.6 times more likely to be able to accurately estimate the cost to operate software. They are also twice as likely to be able to easily identify their most operationally expensive applications.

For example, many organizations with cloud infrastructure do not allow developers to self-service their environments on demand: they require them to raise tickets or send emails and wait days for environments to be provisioned or for configuration changes to be made. In this case, although the organization may pay for a cloud service, they don't have a cloud by NIST's definition.

When moving to the cloud, organizations must invest in re-designing the processes and practices that they implemented when they were using traditional data centers. They must adopt cloud-native patterns and practices to achieve the agility, stability, availability, and cost transparency of cloud computing. If the underlying technology is in the cloud but the outcomes for practitioners remain unchanged — days or weeks to obtain test environments, provision new infrastructure, or get configuration changes made — then organizations will not reap the results they want.

How to implement cloud infrastructure

Adopting cloud-native processes and practices to implement NIST's five characteristics involves substantial change by several IT functions, including developers, operations teams, information security, procurement, and finance. The changes require close collaboration between these functions in order to identify and resolve conflicting aims.

For example, many developers expect complete control over production infrastructure when using cloud platforms. Information security professionals oppose this practice, and with good reason — without disciplined change management, cloud infrastructure can turn into a fragile work of art that is hard to manage, vulnerable to external threats, and does not meet regulatory requirements.

However, developers can be empowered to provision the resources they require while meeting these requirements. Many organizations have adopted the infrastructure as code paradigm. (GitOps is an example.) Infrastructure configuration is checked into version control, and developers can provision environments, make configuration changes, and execute deployments through an automated mechanism. The mechanism takes the code from version control and performs operations through the cloud's API on demand without human intervention. Using version control and automation, all actions and their output are logged to provide complete traceability and auditability of every change to each environment.

Infrastructure-as-code allows you to manage changes effectively, and to apply information security controls. For example, you can implement segregation of duties by requiring all changes to the configuration specified in version control to be approved by someone from a specified group of people (as is done at Google). However, moving to Infrastructure-as-code requires significant engineering effort and process change, including changing policies for implementing information security controls.

Consider another example. While cloud providers typically only bill for infrastructure while it is in use (meeting NIST's fifth characteristic, measured service), this change from fixed-cost infrastructure to variable cost can have significant implications for both procurement and finance. As described in the 2019 State of DevOps Report:

"Some in finance may say that the cloud has not led to cost savings in the short-run, yet we know that it provides greater information transparency. How can this be? While the cloud provides transparent information about costs to the system owners, users do not pay for these costs unless there is a chargeback model or similar mechanism. This can lead to wildly variable costs that go unchecked, making cloud costs unpredictable. In these scenarios, teams that pay for infrastructure may prefer data centers because they are predictable, even though their visibility disincentivizes system users to build more efficient systems. We suggest organizations better align incentives so that system owners have both visibility to build more efficient systems, and the incentives to do so, by using chargeback or similar mechanisms."

In addition to considering how infrastructure is managed at both the configuration and finance level, applications must often be re-architected to reap the benefits of increased stability, reliability, and agility that cloud can provide. Re-architecting systems to a cloud-native paradigm is discussed in Google Cloud's white paper, Re-architecting to cloud native: an evolutionary approach to increasing developer productivity at scale.

The most crucial consideration is whether your cloud deployment has actually helped your organization achieve more rapid, reliable releases, and higher levels of availability, velocity, and reliability.

Common pitfalls of implementing cloud infrastructure

The biggest obstacles to meeting the five characteristics defined by NIST are:

  • Insufficient alignment and collaboration between the organizational functions that must work together in order to implement them;
  • Insufficient investment in technical, process, and organizational change.

Many organizations begin with a lift and shift approach to moving applications to the cloud. This requires minimal change to applications and to established processes for managing cloud infrastructure, and can be done relatively quickly. However, it also provides minimal benefits. It's important to plan for moving beyond lift and shift to a cloud-native paradigm for new software, as well as existing strategic systems that will continue to evolve.

Moving to the cloud is a multi-year journey. Like all disruptive changes, it's important to start small and experiment rapidly to discover what works and what doesn't, and then be persistent and disciplined about scaling out learnings and effective patterns and practices through the organization.

There will be many obstacles on this journey to overcome, including:

  • Redesigning processes, architecture, and governance to meet regulatory requirements in a cloud-native way
  • Designing multi-tenant infrastructure architecture that enables teams to self-service deployments and configuration changes while ensuring logical separation between environments, enabling charge-back, and preventing cloud sprawl and orphaned infrastructure
  • Building a product development capability for your cloud infrastructure platform
  • Helping procurement transition to infrastructure as a metered service rather than a capital investment
  • Helping developers understand how to build cloud-native applications
  • Enabling operations to move to modern Site Reliability Engineering (SRE) practices, including deploying infrastructure-as-code as a replacement for manual ticket-based configuration management
  • Planning and executing the integration between cloud native systems and non-cloud-based systems, including mainframes and packaged software/COTS

Overcoming these obstacles requires a substantial change program involving sustained investment and collaboration between people at every level of the organization.

How to measure cloud infrastructure

In order to decide what to measure, start by considering what benefits you hope to gain from implementing cloud infrastructure. Then start measuring the extent to which those benefits are being realized.

For example, if your goal is to improve cost visibility, you might track how well you're doing at making sure the cost of infrastructure is correctly billed to the relevant business line, team, product, or environment.

You can also directly measure whether or not you have done a good job of implementing NIST's essential characteristics: ask your teams about the extent to which they agree with the statements of these characteristics listed at the top of this document. Teams that agree or strongly agree are doing well; teams that are neutral or disagree need help and support to remove obstacles to meeting these outcomes. Then consider how many teams that say they are using cloud can actually meet NIST's criteria.

Some aspects of the essential characteristics can also be measured directly by instrumenting your processes. For example, if you have a process for managing infrastructure changes you could measure the time it takes end-to-end to make a change. You can also look at how prevalent cloud-native technologies are in your organization: for example, the proportion of clusters or machines managed using Kubernetes or autoscaling groups rather than manual provisioning, or the time-to-live for hosts. In data centers, long uptimes generally indicate high reliability; in the cloud-native paradigm, configuration changes are often made by starting new virtual hosts with the new configuration rather than making changes to existing hosts. This practice is known as ephemeral infrastructure, in which a long uptime is an indicator of an unpatched machine.

What's next