CANDISE: Change-Oriented Assessments for Net-Zero Digital Services

随着气候变化的影响在全球范围内显现，立即减少碳排放至关重要。英国的目标是到2035年减少78%的排放量，其中2%到4%来自ICT行业。联合国ICT行业机构（国际电联）设定了到2030年实现45%的脱碳目标，而对ICT服务的需求也在不断增长，因为人们越来越多地采用新的、更好的服务。运营数据中心和网络的信息通信技术公司，以及那些在物理互联网基础设施之上提供数字软件服务的公司，开始设定自己的净零碳减排目标。现在，他们都需要根据具体变化，通过具体的减排策略来实现这些目标。然而，软件服务的哪些变化将可靠地导致碳减排的知识是缺失的，因为现有的理解忽略了数字服务背后复杂的、高度动态的ICT供应链之间的相互作用。目前，会计角度隐含在对当前碳影响的推理和估计中，这是决定如何减少碳排放的基础。这是一个严重的问题。本项目旨在创造缺失的因果关系理解，并研究一个框架，以预测数字服务的设计、提供和使用变化对系统、长期环境的影响，并为我们追求净零目标的能力提供一步一步的改进。我们的项目团队代表了构成ICT产品系统的整个垂直技术堆栈，从面向用户的数字服务到网络和数据中心运营商。我们有经验和能力来研究用户、软件供应商和基础设施运营商的决策对长期碳影响的原因和评估方法。让我们考虑一下仅仅计算现有的碳排放量和估计未来火车旅行行动的减少量之间的区别。粗略地说，使用一种会计方法，每位乘客每英里的燃料足迹是整列火车每英里的燃料消耗除以平均乘客数量（该方法与火车和轨道建设的影响类似）。重要的是，这样的每乘客每英里的碳足迹不能用来估计每个不旅行的乘客可以节省的碳量。相反，脱碳火车旅行需要长期的改变，比如修改燃料类型、火车结构和时间表。由于乘客不能根据燃料类型自由选择火车，因此向消费者提供这些信息可能是可以接受的。然而，软件供应商在考虑他们的净零计划时，不能使用模拟会计风格的推理来估计云计算和网络的碳足迹，因为他们会错误估计他们的计划的长期后果。与火车乘客相比，软件提供商通过选择架构和供应商，在设计和运营服务方面拥有一定的自由。它们塑造了对基础设施峰值性能的需求，并与即时和持续可用性相结合，因此与云计算和网络共同发展。因此，软件工程师和设计师需要了解他们的决策是如何长期推动碳排放的——这样他们就知道，例如，降低视频分辨率能节省多少碳排放？非峰值预加载内容的真正碳效益是什么？我们与运营着一些世界上最大的ICT服务的合作伙伴一起，研究在软件设计、运营和使用过程中所采取的选择是如何影响环境的。在此基础上，我们研究一个可以支持设计人员和开发人员的框架。这将从系统的、因果关系的理解中释放创新，并为我们追求净零目标的能力提供一步改进。

With discernible effects of climate change manifesting globally, immediate reduction of carbon emissions is crucial. The UK aims to reduce emissions by 78% by 2035, of which 2% to 4% come from the ICT sector. The UN sector body for ICT (ITU) has set decarbonisation goals of 45% by 2030 - while demand for ICT services continues to grow from increased uptake of new and better services.The ICT companies that operate datacentres and networks, and those companies that provide digital software services 'on-top' of the physical internet infrastructure begin to set their own Net Zero goals of carbon emission reductions. Now they each need to implement these goals through concrete reduction strategies based on specific changes.However, the knowledge of what changes to software services will reliably result in carbon reductions is missing, as the existing understanding ignores the interactions between the complex, highly dynamic ICT supply chains underlying digital services. Currently, an accounting perspective is implicit in reasoning about and estimating current carbon impacts and which forms the basis for deciding how to reduce carbon. This is deeply problematic.This project aims to create the missing causal understanding and investigate a framework for anticipating the systemic, long-term environmental impact from changes to the design, provision, and use of digital services and provide a step-improvement to our ability to pursue Net-Zero goals. Our project team represents the entire vertical stack of technologies that form the ICT product system, from user-facing digital services to network and datacentre operators. We have the experience and capacity to investigate ways to reason about and assess the longer-term carbon impact from decisions taken by users, software providers, and infrastructure operators. Let's consider the difference between merely accounting for existing carbon emissions and estimating future reductions from actions in the case of train travel. Roughly, using an accounting method, the footprint per passenger per mile from fuel is the per-mile fuel consumption of the entire train divided by the average number of passengers (the approach is similar for impact from construction of trains and tracks).Importantly, such a carbon footprint per passenger-mile cannot be used to estimate the amount of carbon that could be saved per passenger that was not travelling. Instead, decarbonising train travel requires long-term changes, such as modifying fuel types, train construction, and schedules. As passengers are not free to choose between trains by their fuel types, it might be acceptable to provide this information to consumers. However, software providers must not use analogue accounting-style reasoning to estimate carbon footprints for cloud computing and networks when thinking about their Net Zero plans, as they would misestimate the longer-term consequences of their initiatives. In contrast to train passengers, software providers have a certain freedom in the design and operation of their services through their choice of architecture and suppliers. They shape demand for peak performance of infrastructure paired with instant and constant availability and so co-evolve with cloud computing and networks.Software engineers and designers thus need an understanding of how their decisions drive carbon emissions long-term - so they know e.g. What carbon savings come from reducing video resolution? What is the real carbon benefit from pre-loading content off-peak? Etc.With our partners that operate some of the world's largest ICT services, we study how environmental impact is driven by choices taken during software design, operation and use. From this, we investigate a framework that can support designers and developers. This will unlock innovation from a systemic, causal understanding and provide a step-improvement to our ability to pursue Net-Zero goals.