February 26, 2020

137 words 1 min read

Towards Hyperscale High Performance Computing with RDMA

Towards Hyperscale High Performance Computing with RDMA

RDMA is increasingly deployed in datacenter environments. Deployment introduces an additional level of fabric complexity in management and diagnosis. This talk wil …
Talk Title Towards Hyperscale High Performance Computing with RDMA
Speakers Omar Cardona, Microsoft
Conference NANOG76
Conf Tag
Location Washington DC
Date Jun 10 2019 - Jun 12 2019
URL Talk Page
Slides Talk Slides
Video Talk Video

RDMA is increasingly deployed in datacenter environments. Deployment introduces an additional level of fabric complexity in management and diagnosis. This talk will introduce operators to RDMA concepts from a Fabric management point of view, inclusive of caveats, quirks, misconceptions which have skewed deployment decisions. With multiple variants of RDMA at play (IB, iWARP, RoCE) and differing qualities/properties of each, from workload level capabilities to E2E fabric resiliency; a thorough understanding of each is needed for successful deployment.

management complexity hyperscale performance datacenter
comments powered by Disqus
Deep learning at scale: Tools and solutions

Deep learning at scale: Tools and solutions

February 6, 2020

Success with DL requires more than just TensorFlow or PyTorch. Angela Wu, Sidney Wijngaarde, Shiyuan Zhu, and Vishnu Mohan detail practical problems faced by practitioners and the software tools and techniques you'll need to address the problems, including data prep, GPU scheduling, hyperparameter tuning, distributed training, metrics management, deployment, mobile and edge optimization, and more.
The ultimate guide to complicated systems

The ultimate guide to complicated systems

February 25, 2020

Building and maintaining distributed systems is hard. Industry tools and recommended practices are evolving at an ever-increasing velocity. New platform choices reduce infrastructure management and add operational complexity obscuring the value of operation skills. Often, bureaucratic decisions drive practices and tool choices.
Best Practices and Lessons Learned for Running Kubernetes on Bare-Metal at Scale

Best Practices and Lessons Learned for Running Kubernetes on Bare-Metal at Scale

February 22, 2020

Platform9 offers a SaaS hybrid cloud solution, managing +300 cloud regions & thousands of nodes for our users, globally. Our offering is built on top of Kubernetes on bare-metal. When running Kubernet …
Executive Briefing: The black boxInterpretability, reproducibility, and data management

Executive Briefing: The black boxInterpretability, reproducibility, and data management

February 19, 2020

The growing complexity of data science leads to black box solutions that few people in an organization understand. Mark Madsen explains why reproducibilitythe ability to get the same results given the same informationis a key element to build trust and grow data science use. And one of the foundational elements of reproducibility (and successful ML projects) is data management.
Apache Hadoop 3.x state of the union and upgrade guidance

Apache Hadoop 3.x state of the union and upgrade guidance

February 16, 2020

Wangda Tan and Wei-Chiu Chuang outline the current status of Apache Hadoop community and dive into present and future of Hadoop 3.x. You'll get a peak at new features like erasure coding, GPU support, NameNode federation, Docker, long-running services support, powerful container placement constraints, data node disk balancing, etc. And they walk you through upgrade guidance from 2.x to 3.x.
Hands-on machine learning with Kafka-based streaming pipelines

Hands-on machine learning with Kafka-based streaming pipelines

February 13, 2020

Boris Lublinsky and Dean Wampler examine ML use in streaming data pipelines, how to do periodic model retraining, and low-latency scoring in live streams. Learn about Kafka as the data backplane, the pros and cons of microservices versus systems like Spark and Flink, tips for TensorFlow and SparkML, performance considerations, metadata tracking, and more.