Avoiding Accelerated-Analytics Antipatterns in the Cloud
, Google
, NVIDIA
Hosted cloud services support frameworks like Apache Spark, RAPIDS, and Dask, as well as scheduling jobs on a range of NVIDIA GPUs, and are a great way to get started accelerating analytic processing. However, it can be challenging to optimize these workloads’ performance. With new possibilities for acceleration come new questions about best practices, capacity planning, and tuning. We'll help you navigate this new and exciting landscape to get the most out of GPU-accelerated analytic processing. You’ll learn what kinds of analytic workloads work especially well on GPUs, how to understand the performance of contemporary analytic frameworks on GPUs with open-source tools, and how to tune your cluster for optimal performance. We’ll back up our recommendations with real application case studies. While we’ll show these techniques and recommendations in the context of Google Cloud Dataproc, you’ll learn something you can use no matter where you’re running analytics on GPUs.