Dmitrii Ustiugov focuses on the intersection of machine learning and systems architecture, including designing fast, scalable, and resource-efficient cloud systems. His research is primarily focused on cloud computing and serverless computing architectures, as well as system support for large-scale LLM inference services.
With the acceleration of digital transformation, enterprises increasingly need to build and deploy applications flexibly and efficiently. The emergence of serverless computing architecture improves developer productivity by fully managing cloud infrastructure, allowing developers to focus on business logic and code. However, serverless cloud systems raise new issues and optimization opportunities that require re-visiting the software-hardware stack. Unfortunately, the reliance of most serverless providers on proprietary infrastructure significantly complicates and slows down innovation in this area.
With his colleagues at the University of Edinburgh and ETH Zurich, Dmitrii’s team has developed and maintained vHive, a full-stack open-source ecosystem for serverless cloud benchmarking, experimentation, and innovation to enable unobstructed research in serverless systems. vHive integrates production-grade components, including AWS Firecracker, Containerd, and Kubernetes, unlocking cross-stack systems research and innovation in realistic cloud systems.
Today, more than 30 universities (including Stanford and MIT) and global companies use vHive, such as Intel, to evaluate and improve their commercial products, cloud services, and academic project-based education. With the increasing reliance of enterprises, research institutions, and governments on cloud computing, the importance of research in cloud systems is self-evident. Dmitrii’s ambition is to confront and surmount cloud infrastructure's fundamental challenges, including scalability and energy efficiency.
