5 papers
Tangram: Hiding GPU Heterogeneity for Efficient LLM Parallelization
Yanda Tao, Pedro F. Silvestre, Marcel Wagenländer +1
The scale of LLM training jobs requires parallelization planning over large GPU clusters. Due to different GPU types and interconnects added over time, these GPU clusters are incre…
Scepsy: Serving Agentic Workflows Using Aggregate LLM Pipelines
Marcel Wagenländer, Otto White, Britannio Jarrett +6
Agentic workflows carry out complex tasks by orchestrating multiple large language models (LLMs) and tools. Serving such workflows at a target throughput with low latency is challe…
Nanvix: A Multikernel OS Design for High-Density Serverless Deployments
Carlos Segarra, Pedro Henrique Penna, Enrique Saurez +4
Serverless providers strive for high resource utilization by optimizing deployment density: how many applications can be deployed per host server. However, achieving high deploymen…
Cloud abstractions for AI workloads
Marco Canini, Theophilus A. Benson, Ricardo Bianchini +4
AI workloads, often hosted in multi-tenant cloud environments, require vast computational resources but suffer inefficiencies due to limited tenant-provider coordination. Tenants l…
ARCAS: Adaptive Runtime System for Chiplet-Aware Scheduling
Alessandro Fogli, Bo Zhao, Peter Pietzuch +1
The growing disparity between CPU core counts and available memory bandwidth has intensified memory contention in servers. This particularly affects highly parallelizable applicati…