Systems for AI and Big Data

Sustainable AI Systems for Energy-Efficient Inference

In PINet lab, we design sustainable AI systems to address the energy challenges of inference in cloud data centers and edge environments. By combining system-level optimizations, such as dynamic resource management, with model-level strategies like parameter tuning and workload adaptation, we create innovative solutions that balance energy efficiency, performance, and scalability. Our work aims to enable the next generation of AI technologies that are both environmentally sustainable and impactful across diverse applications.

Rapid Prototyping of Stream Processing Applications

Stream processing is becoming mainstream. It is present in thousands of companies around the globe and used in a variety of applications including real-time monitoring, data analytics, car-trip pricing, credit card fraud detection, and many others. Despite the great success of the stream processing paradigm, testing a networked stream processing application (particularly at scale) is still a cumbersome and often expensive process. Existing approaches typically include building an expensive testbed, applying for a slot on a community one (e.g., GENI, CloudLab) or hiring a cloud-based set-up. On top of that, developers still need to spend significant amounts of time configuring and managing their platforms, e.g., by connecting cluster nodes and ingesting data. In this project, we aim at lowering the wall (from both time and money perspective) for stakeholders to participate in the stream processing ecosystem. For that, we are building a simple and easy-to-use prototyping tool for distributed stream processing applications. It is a whole stream processing setup in a single server and in a matter of minutes. We envision this work can also help paving the way for more reproducible research in the stream processing domain, a basic need for the research community.


Publications

GreenStream: Enabling Sustainable LLM Inference in Stream Processing
Md. Monzurul Amin Ifath, Israat Haque
Won the Best-Poster Runner Up Award at CASCON 2024

[Poster]

Automated Configuration Parameter Tuning in Distributed Messaging Systems
Emmanuel Etti, Md. Monzurul Amin Ifath, Israat Haque
Accepted for poster presentation at CASCON 2024

[Poster]

Are data streaming platforms ready for a mission critical world?
Md. Monzurul Amin Ifath, Miguel Neves, Brandon Bremner, Jeff White, Tomas Szeredi, Israat Haque
Submitted at IEEE Communications Magagine 2024

[Paper]

Fast Prototyping of Distributed Stream Processing Applications with stream2gym
Md. Monzurul Amin Ifath, Miguel Neves, Israat Haque
IEEE ICDCS 2023

[Paper] [Code] [Bib]

Raptor: Rapid prototyping of distributed stream processing applications at scale
Md. Monzurul Amin Ifath, Miguel Neves, Israat Haque
ACM CoNEXT Posters and Demos 2021

[Paper] [Bib]


People

  • Israat Haque, Dalhousie University, israat at dal dot ca
  • Monzurul Ifath, Dalhousie University, monzurul dot amin at dal dot ca
  • Emmanuel Etti, Dalhousie University, em244217 at dal dot ca
  • Miguel Neves (Past Member), Dalhousie University, mg478789 at dal dot ca