Graphs and GraphSys

The interoperability and analytical exploitation of graph data have become foundational pillars of modern digital economies and scientific discovery. Today, while computational methods and FAIR (Findable, Accessible, Interoperable, and Reusable) graph datasets exist, current computational capabilities are increasingly lagging behind the extreme scale of datasets and the complexity of modern workflows. Historically, performance optimization in graph systems has relied on correlation-based methods, which do not actually model the underlying causal mechanism. As a result, these methods can only identify factors that co-occur with poor performance, but do not provide insights into why this is happening. This effectively limits the ability of these approaches to guide architectural decisions. We argue that the community must move toward reasoning approaches that exploit temporal and causal graph representations to understand and model the fundamental mechanisms that influence outcomes. GraphSys-2026 aims to pioneer the integration of Causal and Temporal Graphs into the system design lifecycle. By treating system telemetry as information that can be translated into a temporal causal structure, we can derive mechanistic insights into hardware phenomena such as memory bottlenecks, energy spikes, and quantum decoherence.

Graphs serve as universal abstractions that capture the interconnectedness of real and digital worlds. By extending these abstractions to represent the causal dependencies between hardware states and software execution, we enable a deeper understanding that can serve a broad range of purposes, from hardware co-design to runtime optimizations. This evolution contributes to the United Nations Sustainable Development Goals (UN SDG), specifically SDG 7 (Affordable and Clean Energy) and SDG 9 (Industry, Innovation, and Infrastructure).

Graphs are getting larger and more dynamic, and applications using them are getting increasingly complex. We argue that the only viable path to making graph processing easy to use, fast, scalable, and sustainable is through cross-disciplinary convergence of parallel computing, architecture co-design, and causal artificial intelligence. GraphSys ‘26 provides the rigorous venue required for specialists to discuss this new state of the art, from the physics of the underlying hardware to the high-level causal reasoning of the algorithms.

Workshop format

GraphSys-2026 will be a full-day (in-person) workshop, running a single track (i.e., no parallel sessions). Central to this edition are three high-impact challenges: (i) causal discovery in HPC and datacenter management, (ii) causal understanding of quantum uncertainty, and (iii) causal hardware–software co-design. We invite submissions that push the boundaries of the field - whether through foundational vision papers, the release of novel datasets, benchmarks, or rigorous academic and industrial advancements addressing our three core challenges. We plan to feature one keynote, 6–10 paper presentations, flash presentations and posters for each challenge, and one panel. We will invite submissions of full papers (12 pages), datasets, work-in-progress, experience, and position papers (6 pages), and extended abstracts addressing the challenges (2 pages). We will run a thorough revision process to select the papers featured in the proceedings. Workshop proceedings will be published in the joint Springer LNCS volume for Euro-Par 2026 workshops.

Sponsorship

Sponsorship opportunities for GraphSys 2026 will be announced soon.