People
- Jeffrey Young
- Alex Merritt
- Se Hoon Shon
Sponsors
- Custom HTX blades supplied by AIC, Inc.
- LogicBlox, Inc.
- NVIDIA
- Intel Science and Technology Center for Cloud Computing (ISTC-CC)
Collaborators
- Computer Architecture Group at University of Heidelberg (Ulrich Brüning, Holger Fröning)
- Parallel Architectures Group (GAP) at Polytechnic University of Valencia (José Duato, Federico Silla)
Oncilla: A GAS Runtime for Memory Allocation and Data Movement in Clusters
Oncilla is a project that aims to provide a commodity-based, non-coherent managed global address space (GAS) runtime to support efficient data movement between host memory (DRAM) and accelerators (GPUs). Oncilla provides basic allocation and aggregation support for applications as well as support for multiple high-performance networking fabrics. Currently we support two network fabrics – the University of Heidelberg’s EXTOLL fabric and InfiniBand Verbs. Allocations and data movement are exposed to applications through the Oncilla API – a simplified interface built into a Linux shared library.
Current requirements for using Oncilla include the use of either IB (via the OFED software stack) or EXTOLL networking stacks (RMA2) and the use of CUDA for GPU allocations. Oncilla is currently hosted at Github, although please be advised that this is alpha research code. For more information on the Oncilla project, please refer to our IEEE Cluster paper that can be found in the list of publications.
Oncilla supported data paths for Red Fox
Publications
J. Young, S. Shon, S. Yalamanchili, A. Merritt, K. Schwan, H. Fröning, Oncilla: A GAS Runtime for Efficient Resource Allocation and Data Movement in Accelerated Clusters. IEEE Cluster. September 2013. paper
H. Wu, J. Young, and S.Yalamanchili. “Satisfying Data-Intensive Queries Using GPU Clusters (Poster).” GPGPU Technology Conference (GTC). March 2013. poster
J. Young, A. Merritt, H. Wu, and S. Yalamanchili. “Oncilla – A GAS Run-time for Efficient Resource Partitioning in Data Centers (Poster).” 2nd Annual Intel Science and Technology Center for Cloud Computing (ISTC-CC) Retreat. December 2012. poster
J. Young, H. Wu, and S. Yalamanchili. “Satisfying Data-Intensive Queries Using GPU Clusters.” 2nd Annual Workshop on High-Performance Computing meets Databases (HPCDB), held with SC12. November 2012. paper and slides
S. Yalamanchili, J. Young, and A. Lele. “Oncilla – Optimizing Accelerator Clouds for Data Warehousing Applications.” CASL Whilte Paper. 2012. Oncilla White Paper
The papers are provided for personal use and are subject to copyright of the publishers.
