Chameleon@SC21

Attending SC21? Check out the initiatives Chameleon is supporting including workshops, papers written using Chameleon, and more! If you’re presenting something at SC21 and not on this list, send us an email! 

 

Learn about Chameleon

Information about Chameleon will be presented in the StarLight booth (2727 next to the SCinet booth) at SC21, by Chameleon partner organization, the International Center for Advanced Internet Research at Northwestern University.

 

IndySCC 2021 on Chameleon@NCAR

The 2021 Indy Student Cluster Competition (IndySCC) asks teams of students to optimize the performance of several classes of computation, while staying beneath a strict power budget. This year, Chameleon is hosting the competition, with the teams using a variety of nodes to complete assignments, finishing with their final code evaluated during a 48 hour benchmark run on November 6th and 7th.

In addition to existing GPU nodes in CHI@TACC, the National Center for Atmospheric Research (NCAR) and Chameleon have paired up to offer five ARM ThunderX2 nodes, purchased as test nodes by the National Science Foundation, to Chameleon users as a Chameleon Associate site. This has allowed a new format for the competition, where the teams will need to pick the best combination of two P100 GPU nodes and one of these ARM nodes to maximize their performance per watt.

Learn more about Chameleon supporting the IndySCC at a lightning talk given by Michael Sherman, University of Chicago site lead, during the HPCSYSPROS SC21 workshop on Sunday, November 14, from 3:30p-3:45p CST. The full HPCSYSPROS schedule is available here.

 

SC Reproducibility Initiative

To encourage the reproducibility of scientific research, Supercomputing 2021 is hosting a Reproducibility Initiative. Authors submitting papers are encouraged to make artifacts available, allowing reviewers and readers to reproduce the experiment and validate the results. Chameleon has served as an evaluation platform for this initiative and has provided webinars and training to authors and reviewers alike, allowing the review committee to perform single-blind reviews of these artifacts. The report of the SC21 Reproducibility inititative can be read here.


 

Papers authored by the Chameleon Community: 

  1. NetGraf: An End-to-End Learning Network Monitoring Service, Bashir Mohammed, Mariam Kiran, Bjoern Enders, Lawrence Berkeley National Laboratory, Berkeley California. USA

 

  1. Cuttlefish: Library for Achieving Energy Efficiency in Multicore Parallel Programs. Sunil Kumar, Akshat Gupta, Vivek Kumar, and Sridutt Bhalachandra. In Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC '21). To appear. Association for Computing Machinery, New York, NY, USA. [Link]

  2. FedAT: a high-performance and communication-efficient federated learning system with asynchronous tiers. Zheng Chai, Yujing Chen, Ali Anwar, Liang Zhao, Yue Cheng, and Huzefa Rangwala. In Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis. Association for Computing Machinery, New York, NY, USA, Article 60, 1–16. DOI:https://doi.org/10.1145/3458817.3476211 [link]

 

Posters authored by the Chameleon Team:

  1. One Fish, Two Fish: Choosing Optimal Edge Topologies for Real-Time Autonomous Fish Surveys. Tsen, J., Anderson, J.,  Bobadilla, L. and Keahey K., In Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC’21 Poster). IEEE Press. November 2021

 

  1. Yin and Yang: Balancing Cloud Computing and HTC Workloads. Kang Z., Zhen Z., and Keahey K., In Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC’21 Poster). IEEE Press. November 2021

 

Demos

Two large scale National Science Foundation computer science testbed projects have been planning to provide integrated resources for their communities: Chameleon, a large-scale, deeply reconfigurable experimental platform for Computer Sciences systems research, and FABRIC, which enables cutting-edge and exploratory research at-scale in networking, cybersecurity, distributed computing and storage systems, machine learning, and science applications. Currently these projects are investigating methods of optimizing cross platform research.  

As a first step, these projects are designing demonstrations that will be staged at the IEEE/ACM International Conference For High Performance Computing, Networking, Storage, and Analytics, November 14-19, 2021, in St Louis, Mo, which annually showcases breakthrough discoveries and innovations. Demonstrations being considered are experiments using Jupyter notebooks to integrate Chameleon and FABRIC resources. These notebooks could be shared by being published with Trovi. Also being considered is implementing an L2 stitched network between Chameleon and FABRIC and using it from slices deployed with a single Jupyter notebook. These demonstrations may involve SCinet, the advanced high-capacity network designed specifically for SC conferences. The details of these demonstrations and their results will be broadcast after the conference.

 

SCinet 

Chameleon is also providing a development and testbed platform for SCinet. 


Add a comment

No comments