Power- and Fragmentation-aware Online Scheduling for GPU Datacenters

The rise of Artificial Intelligence and Large Language Models is driving increased GPU usage in data centers for complex training and inference tasks, impacting operational costs, energy demands, and the environmental footprint of large-scale computing infrastructures. This work addresses the online scheduling problem in GPU datacenters, which involves scheduling tasks without knowledge of their future arrivals. We focus on two objectives: minimizing GPU fragmentation and reducing power consumption. GPU fragmentation occurs when partial GPU allocations hinder the efficient use of remaining resources, especially as the datacenter nears full capacity. A recent scheduling policy, Fragmentation Gradient Descent (FGD), leverages a fragmentation metric to address this issue. Reducing power consumption is also crucial due to the significant power demands of GPUs. To this end, we propose PWR, a novel scheduling policy to minimize power usage by selecting power-efficient GPU and CPU combinations. This involves a simplified model for measuring power consumption integrated into a Kubernetes score plugin. Through an extensive experimental evaluation in a simulated cluster, we show how PWR, when combined with FGD, achieves a balanced trade-off between reducing power consumption and minimizing GPU fragmentation.

Urban Region Embeddings from Service-Specific Mobile Traffic Data

With the advent of advanced 4G/5G mobile networks, mobile phone data collected by operators now includes detailed, service-specific traffic information with high spatio-temporal resolution. In this paper, we leverage this type of data to explore its potential for generating high-quality representations of urban regions. To achieve this, we present a methodology for creating urban region embeddings from service-specific mobile traffic data, employing a temporal convolutional network-based autoencoder, transformers, and learnable weighted sum models to capture key urban features. In the extensive experimental evaluation conducted using a real-world dataset, we demonstrate that the embeddings generated by our methodology effectively capture urban characteristics. Specifically, our embeddings are compared against those of a state-of-the-art competitor across two downstream tasks. Additionally, through clustering techniques, we investigate how well the embeddings produced by our methodology capture the temporal dynamics and characteristics of the underlying urban regions. Overall, this work highlights the potential of service-specific mobile traffic data for urban research and emphasizes the importance of making such data accessible to support public innovation.