Distributed and Operating Systems Research at TU Berlin

Distributed and Operating Systems is a research group at TU Berlin by professor Odej Kao. We work at the intersection of distributed systems, operating systems, artificial intelligence, and information systems.

Research

We design and evaluate systems and methods for operating modern data-intensive and AI-driven applications efficiently, reliably, and sustainably across cluster and cloud infrastructures. Our current research topics cover

🧠 Systems for AI

We build architectures that optimize how AI operates within distributed environments, focusing on making workloads faster, more efficient, and safer. Our core structural research includes distributed AI, adaptive cluster computing, and Mixture of Experts. To accelerate performance, we design mechanisms for efficient inference, inference optimization for speedups, and predicting model performance for foundation models. We also investigate the orchestration of intelligent systems through AI agent orchestration routing and AI-defined processes. Furthermore, we address AI safety and capabilities by exploring AI security in non-LLM contexts, activation monitoring, model interpretability, AI finetuning, reinforcement learning, Chain-of-Thought length control, and graph deep learning.

🕵 Observability

Modern AI platforms must remain robust under failures and anomalies. We develop methods and systems to ensure reliable AI and robust deep learning. Our work encompasses comprehensive observability and fault tolerance mechanisms to make training and inference pipelines resilient. This involves advanced anomaly detection and developing methods to specifically identify and mitigate silent data corruption within distributed operations.

⚙️ Operation

Managing and executing distributed workloads requires robust operational strategies, with a strong focus on sustainable computing. We develop approaches for dynamic resource management and modeling uncertainty to optimize how tasks are placed and executed across heterogeneous infrastructures. A core part of our operational research targets the ecological footprint of these systems through energy system modeling and carbon accounting. We also explore innovative operational incentives, such as utilizing energy as a reward metric, to ensure that large-scale infrastructure is run both highly efficiently and sustainably.