Glasgow Carbon-Conscious Computing (GC3) lab

The GC3 lab is a new computer systems laboratory in the Glasgow Systems Section and the School of Computing Science at the University of Glasgow. Its research falls into the School's newest research theme, Low-Carbon and Sustainable Computing.

Interested in joining as a PhD student? We are looking for motivated students to work with us on sustainable computer systems.

GC_3 Lab Logo
starting point:

There are increasingly many applications that make use of large amounts of data. This includes scientific data processing in research organizations as well as big data applications in industry. Meanwhile, there are also more and more data-intensive applications in context of the distributed sensors and devices of the Internet of Things.
We believe there is immense potential in using data-driven methods and machine learning across domains and disciplines. Though, if we are not careful, this development will further increase computing’s environmental footprint. This footprint is already estimated to contribute 2 to 3 percent of the global carbon emissions – very much rivaling aviation – and is projected to dramatically rise further over the next decades, not least because of the trend towards more data-intensive applications.

research agenda:

The central aim of our research is to reduce the carbon footprint of large-scale resource-intensive systems on today's diverse computing infrastructures. To this end, we work on methods and tools that support a more resource-efficient and sustainable use of modern distributed computing infrastructures (e.g. edge and cloud resources) for scalable data-intensive applications (e.g. machine learning, big data analytics, stream processing), while still taking performance and dependability requirements into account.

Our guiding principles for this are:

  1. compute when and where green energy is available (which can, for instance, translate to carbon-aware scheduling, scaling, and resource allocation) as the carbon intensity of grids and renewable energy generation often varies
  2. allocate resources for high resource utilization and highly utilize allocated resources (which can, for example, translate to "right-sizing", server consolidation, co-location of jobs with complementary resource demands, and bottleneck mitigation) as any energy is wasted with mostly idling resources
  3. save computation and communication through distributed and dynamic architectures (which can, for instance, translate to edge computing, adaptively offloading tasks and scaling out to more nodes only when necessary, effective caching, or distributed learning) as it is sometimes possible to do the same with less in complex software systems

Much of the work builds upon years of research on adaptive resource management for data-intensive applications (e.g. for big data analytics, scientific workflows, and machine learning) at and with TU Berlin. On this basis, we currently focus on investigating and making use of:

  • profiling and predicting application and infrastructure performance as well as power consumption
  • forecasting of computational loads, carbon emissions, and the availability of renewable energy
  • carbon-optimized resource allocation, dynamic scheduling and scaling, and automatic system tuning

Another focus is on techniques and tools that support monitoring, testing, and benchmarking of the performance and footprint of data-intensive distributed systems.

team:
  • academic lead: Dr. Lauritz Thamsen
  • postdoc researcher: Dr. Youssef Moawad
  • PhD students: James Nurdin, Kathleen West, Max MacDonald (incoming)
  • L4/L5 students: Andres La Riva Perez, Danial Tariq, Isabella Gard, Magnus Reid, Matthew Waters
  • lab alumni:
    • L4/L5 projects: James Sharma, John Wilson, Karl Hartmann, Richard Arthurs, Rishabh Mathur
    • summer internships: Domonkos Revesz, Frederik Glitzner, Hubert Dymarkowski, Niovi Lampiri
visitors:
selected results:

Several of these works are featured in my talk in our Low-Carbon and Sustainable Computing seminar series and the recording of the talk is available online.

academic collaborations:
industry collaborations:
research infrastructure:

For our research, we have access to CPU and GPU clusters as well as a variety of individual devices (e.g. different server architectures, IoT systems, energy meters) in Glasgow. In addition, we are regularly able to make use of other compute infrastructure (e.g. public/private cloud services) through grants and collaborations.