A low-carbon computing platform from your retired phones
A low-carbon computing platform from your retired phones
利用退役手机构建低碳计算平台
June 12, 2026 Jennifer Switzer, Visiting Postdoctoral Researcher, and David Patterson, Fellow, Google 2026年6月12日 Jennifer Switzer，访问博士后研究员；David Patterson，Google 研究员
With support from Google, Researchers at the University of California San Diego are building a useful second-life for consumer smartphones. The carbon footprint of computing is a key sustainability challenge. It is driven by two major sources: operational carbon reflects emissions from energy consumed during use, and embodied carbon encompasses emissions associated with hardware manufacturing. While operational carbon is often addressed with efforts such as improved energy efficiency and using clean energy, the manufacturing footprint represents a more complex hurdle. 在 Google 的支持下，加州大学圣地亚哥分校的研究人员正在为消费级智能手机创造有价值的“第二生命”。计算产生的碳足迹是可持续发展面临的一项关键挑战。它主要源于两个方面：运营碳排放（反映了使用过程中消耗能源产生的排放）和隐含碳排放（涵盖了与硬件制造相关的排放）。虽然运营碳排放通常可以通过提高能效和使用清洁能源等努力来解决，但制造环节的碳足迹却是一个更为复杂的难题。
To address this, researchers at the University of California San Diego are building a pathway for the second life of phones through the exploration of “phone cluster computing.” This is a process whereby the motherboards of retired smartphones are extracted, collected into clusters, and redeployed as a general-purpose computing platform. With Google’s support, the university plans to deploy a datacenter built from 2,000 Pixel smartphones that will provide hundreds of researchers and students with low-cost, low-carbon cloud computing, reducing the need for newly-manufactured hardware and their associated emissions. 为了解决这一问题，加州大学圣地亚哥分校的研究人员正在通过探索“手机集群计算”为手机的第二次生命开辟路径。这一过程包括提取退役智能手机的主板，将其组装成集群，并重新部署为通用计算平台。在 Google 的支持下，该大学计划部署一个由 2,000 部 Pixel 智能手机组成的数据中心，为数百名研究人员和学生提供低成本、低碳的云计算服务，从而减少对新制造硬件及其相关排放的需求。
Smartphones: A significant contributor
智能手机：重要的贡献者
On average, people replace their phone every four years. This is generally driven by people’s desire for a new device, including for the functionalities provided by new models. Many replaced phones, however, have their core compute functionalities intact and are still relatively powerful computers with integrated processors, accelerators, memory, and storage. While an old phone might no longer be of interest to its first purchaser, putting it back in service can directly reduce the environmental footprint of computing by avoiding the need for further raw material extraction. 平均而言，人们每四年更换一次手机。这通常是由人们对新设备的渴望所驱动，包括对新机型所提供功能的追求。然而，许多被更换的手机其核心计算功能依然完好，它们仍然是集成了处理器、加速器、内存和存储的相对强大的计算机。虽然旧手机可能不再吸引其最初的购买者，但将其重新投入使用可以通过避免进一步的原材料开采，直接减少计算对环境的影响。
This blog discusses a novel strategy: re-deploying unwanted smartphones for cloud computing applications. 本博客讨论了一种新颖的策略：将闲置的智能手机重新部署用于云计算应用。
The single-threaded performance of modern smartphones’ performance processor cores is on-par with or better than those of modern multicore servers (see figure below). The most significant difference between a smartphone and a server is their size: servers contain dozens of powerful multithreaded processor cores and a huge memory capacity, while a smartphone has a handful of heterogeneous processor cores and 8-12GB of memory. One of the key challenges, then, is to target applications that fit into, or can be made to fit into, the capacity of a smartphone. 现代智能手机高性能处理器核心的单线程性能与现代多核服务器相当，甚至更胜一筹（见下图）。智能手机与服务器之间最显著的区别在于规模：服务器包含数十个强大的多线程处理器核心和巨大的内存容量，而智能手机只有少数几个异构处理器核心和 8-12GB 的内存。因此，关键挑战之一是针对那些适合（或可以调整为适合）智能手机容量的应用程序。
From consumer devices to datacenter hardware
从消费设备到数据中心硬件
Redeploying unmodified consumer smartphones in a datacenter environment would be hazardous and inefficient. Smartphones’ compute elements are wrapped in components that aren’t needed in the server context — display, battery, chassis, and peripheral hardware like cameras. In addition to taking up valuable space, some components, such as batteries, contain materials not rated for a datacenter environment. 在数据中心环境中直接重新部署未经改装的消费级智能手机既危险又低效。智能手机的计算元件被包裹在服务器环境下不需要的组件中，例如显示屏、电池、外壳以及摄像头等外围硬件。除了占用宝贵的空间外，某些组件（如电池）还含有不符合数据中心环境要求的材料。
Prior to deployment, smartphones must be processed to remove all but the motherboard, which contains the core compute functionality. Note that the motherboard is responsible for the largest fraction of embodied carbon (approximately 50% based on internal carbon footprinting assessments), so this effort targets the most impactful components. 在部署之前，必须对智能手机进行处理，拆除除主板以外的所有部件，因为主板包含了核心计算功能。值得注意的是，主板占据了隐含碳排放的最大比例（根据内部碳足迹评估约为 50%），因此这项工作针对的是最具影响力的组件。
The Android operating system (OS) is already based on Linux, but the mobile-oriented Android userspace must be replaced with a general-purpose Linux distro. Updating the OS doesn’t just get programmability; it also switches off many of the protections that are important for consumer devices, but unnecessary for cloud computing. For example, phones have a “low memory killer” daemon, which throttles memory-hungry applications. Android 操作系统本身基于 Linux，但必须将面向移动设备的 Android 用户空间替换为通用的 Linux 发行版。更新操作系统不仅是为了获得可编程性，还可以关闭许多对消费设备很重要、但对云计算而言不必要的保护机制。例如，手机中有一个“低内存杀手”（low memory killer）守护进程，它会限制内存占用过高的应用程序。
The challenge of orchestrating jobs across the large number of devices that are needed to meet the performance of a traditional server — SPEC benchmarking results indicate that 25-50 phones equate to a modern server — is addressed by the use of containerized applications managed by Kubernetes. The phones are organized into self-managing clusters of 25-50 devices. 为了达到传统服务器的性能，需要大量设备协同工作，而协调这些任务是一项挑战（SPEC 基准测试结果显示，25-50 部手机相当于一台现代服务器）。这一问题通过使用由 Kubernetes 管理的容器化应用程序来解决。这些手机被组织成 25-50 台设备组成的自管理集群。
Building a low-carbon cloud computing platform
构建低碳云计算平台
At many universities, an abundance of EdTech, grading, and research applications are already being run on the cloud. These applications range from tiny machines for hosting Jupyter notebooks to expensive GPU-based servers for parallel computing classes. The vast majority of these applications are within the capabilities of a single smartphone to host, with the standard grading backend running on small cloud instances such as AWS’ t3.micro (2 vCPU, 1 GB memory). 在许多大学里，大量的教育技术、评分和研究应用程序已经在云端运行。这些应用范围广泛，从托管 Jupyter Notebook 的小型机器到用于并行计算课程的昂贵 GPU 服务器。绝大多数此类应用程序都在单部智能手机的承载能力范围内，标准的评分后端通常运行在 AWS t3.micro（2 vCPU，1 GB 内存）等小型云实例上。
Researchers at the University of California San Diego are planning a 2,000-phone computing cluster to support computer science classes such as Parallel Computation and Systems Programming. Early experiments show that even a moderately-sized cluster of 20 phones is capable of supporting peak submission rates for a 75+ student class, with grading latencies below the default AWS backend. 加州大学圣地亚哥分校的研究人员正在规划一个由 2,000 部手机组成的计算集群，以支持并行计算和系统编程等计算机科学课程。早期实验表明，即使是一个由 20 部手机组成的中等规模集群，也能够支持 75 名以上学生班级的峰值提交率，且评分延迟低于默认的 AWS 后端。