The Importance of Service Design to Sustainability and AI

Sustainability and AI don’t seem like natural companions. Predictions of energy requirements for AI are eye-popping and will make acheiving sustainability targets very difficult. Yet Service Design, referring specifically to IT services, is fundamental to both.

Introduction

The IT and Energy industries are starting to sit up and take note of the expected growth of our data storage requirements. The exponential growth of data will require more and larger data centres and will place increasing demands on energy consumption at a time when there are also large pressures to reduce carbon based energy and carbon emissions.

A Simple IT Sustainability Framework

To help address this issue of growing energy demand, IT has a large part to play. Improving IT efficiency can reduce energy demand and carbon emissions, and also save lots of money.

IT efficiency can be thought about from three main perspectives:

• Infrastructure (or Hardware) efficiency
• Software efficiency
• Data efficiency

Infrastructure Efficiency

Infrastructure efficiency focuses on optimizing resource utilization and reducing wastage. This is the main focus of Sustainable IT at present - move to the cloud, move to Kubernetes, right size your instances, turn off unused infrastructure, recycle old hardware, etc. It is the basis of FinOps, or cloud cost management. It is predominantly focused on Compute Efficiency, which is a significant energy and cost sink in IT operations. Most articles and blogs that discuss IT sustainability focus on compute efficiency and make recommendations based on improving compute efficiency.

Software Defined Networking (SDN) emerges as a key technology in enhancing Network Efficiency. By transitioning from specialized network devices to generic ones, SDN enables a more dynamic and flexible network infrastructure. This shift facilitates virtualization, hardware consolidation, shared infrastructure usage, and optimized network traffic flows, which collectively contribute to improved energy efficiency and reduced operational costs.

Cloud vendors of course will suggest that the easiest way to be more sustainable is to move to the cloud. But many compute workloads are on-premise and many will remain on-premise for the foreseeable future. How do we make on-prem infrastructure more efficient? Virtualisation and containerisation are two viable options of course.

But there are deeper issues. Large enterprises may own 100s or 1000s of on-premise servers and network devices. This infrastructure is notoriously difficult to keep track off. There is often no central inventory, lack of ownership, and out of date inventory lists managed on spreadsheets.

The first step in Inventory rationalisation is to determine who owns the infrastructure. Ownership should never by assigned to an individual, as individuals come and go. Ownership of infrastrcture should be assigned to a product or service. (I use the term “service” to refer to both products and services.) There should be one central golden source of IT inventory data and that should be the organisation’s CMDB (Configuration Management Database). IT Service Management tools will provide CMDB functionality. But ultimately all items in the IT inventory should be assigned to and owned by an IT service. More on this below.

Software Efficiency

Software efficiency seeks to optimize how software is architected, deployed, and run, to make better use of the underlying hardware. This typically involves containerizing workloads and adopting shared tenancy models, where multiple applications run on shared infrastructure, minimizing idle time. The predominant use of platforms like Kubernetes for container orchestration underscores the move towards more efficient, scalable software deployment models.

Transitioning legacy systems to microservices or adopting serverless architectures are strategies aimed at improving software efficiency. While these require upfront investment in refactoring and redesign, they can lead to significant long-term benefits in agility, scalability, and resource utilization.

Web UIs can also be inefficient. Too many websites and apps send large amounts of data (typically javascript) to browsers and mobile devices because networks are fast and devices performant. There are definitiely efficiency gains to be had making apps more efficient.

There are other software efficiency plays that fall under the banner of Green Software, like re-writing code to be more efficient or moving to a more efficient coding language. Although it is difficult to get data that measures such improvements, I would argue that the ROI on software efficiency is lower (worse) than the ROI on infrastructure efficiency. Turning off infrastructure on evenings and weekends is much simpler than refactoring a monolithic architecture, moving to a serverless based architecture, or re-writing your application in Rust, for example.

Data Efficiency

Data efficiency gets less focus than hardware and software efficiency for a number of reasons. 1) we have been told for many years that “Storage is Cheap”, so the IT industry has an “if in doubt, store it” mentality. 2) Data is seen as a valuable resource (like gold or oil) and is therefore hoarded in case there will be some future value.

Yet data is possibly where most IT waste lies. Data that provides little or no value is called dark data. Various estimates say that up to 90% of our data is dark data. Not only is this expensive in terms of costs and carbon emissions, but low/no value data slows query times, increases data ingress and egress charges and processing times, and degrades the output of AI assistants that rely on this data.

So how do we manage dark data? Ideally data with no value should be deleted, and data with little value, such as historic data, should be archived. This sounds simple in theory, but can be very difficult in practice.

• Historic data often has no owner, so no one currently working in the organisation knows if this data is important.
• Lack of data retention policies mean we don’t know if we need to keep data or not, so we keep it by default.
• There is the Fear of Deleting Data (FODD) - you can’t normally get in trouble by leaving data alone, but you can get into all kinds of trouble if you accidentally delete data that was possibly important.

The key factor in managing large amounts of data in enterprises is to make sure all data has an owner. This data owner should NOT be an individual. People come and go and their knowledge about data goes with them. All data within an organisation should be owned by a product or service team. Services are long lasting and the service owners should understand what data is owned and required by their service. Costs of storing the data can then be attributed to the service, and decisions can be made at the service level as to the data costs, benefits and regulatory requirements.

In order to attribute service ownership to all of an organisation’s data, the organisation must be organised around services. What are the organisation’s services? Who owns these services?

The Role of Service Design

Effective management of IT infrastructure, software, and data through service design is essential for aligning IT services with business needs. This includes assigning service ownership and configuring service definitions using ITSM tools like ServiceNow or Jira Service Management. Both AI and sustainability benefit from clean, well-managed data, underscoring the importance of robust IT service design.

Conclusion

As AI's energy demands grow, integrating sustainability into IT service design becomes imperative. By focusing on infrastructure, software, and data efficiency, IT can lead the charge in achieving sustainability goals, proving that AI and sustainability can indeed be compatible through thoughtful design and strategic management.