Innovation Synergy: Ozone, GenAI, Kubernetes Meetup

Title - Innovation Synergy: Ozone, GenAI, Kubernetes Meetup
When - October24th, 4:30 PM to 6:30 PM
Where - Cloudera Santa Clara Office
For people located remote - https://cloudera.zoom.us/j/92621190663

6:30 PM - Pizza/Coke and networking

Ozone Feature Set and Roadmap - Shiv Moorthy (Cloudera)
04:30 PM - 04:45 PM
Description: This talk provides a high-level quick overview of the Apache Ozone feature set and talks about the exciting roadmap items.
What the audience will learn: The audience will learn what is there in Apache Ozone and what innovative feature set is coming up in the project.

***

Ozone + GenAI Applications - Swaminathan Balachandran (Cloudera) 04:45 PM - 05:00 PM
Description: In this talk, we will explore how GenAI applications can be beneficial with the Apache Ozone's unique potential in object storage space. Apache Ozone has a point-in-time snapshot capability. For RAG applications, unstructured data stored in object storages and many RAG pipelines had to build their own CDC mechanisms to understand the knowledge-based mutations and the need to regenerate vectors. With snapshots, figuring out such modifications is at the user's fingertips. Smart SnapDiff can provide exact modification information on buckets and users can easily integrate and re-generate vectors only for modified knowledge base data files. This talk will provide a demo overview of how we can implement and show results in real-time.
What the audience will learn: The audience will understand how Apache Ozone can be helpful with the GenAI RAG application and improve the overall RAG pipeline efficiency. The slick demo provides a clear insight into the overall process. In the end, the audience would be able to understand and develop your RAG app with Apache Ozone.

***

Simplifying Orchestration of GenAI Applications Across Multi-Cluster Kubernetes Environments - Selvi Kadirvel (Elotl)
05:00 PM - 05:20 PM
Description: Running Gen-AI applications at scale is operationally complex - it requires the management of application workloads; infrastructure services, data pipelines and datasets across a multi-cluster, multi-cloud environment. In this talk, we describe how to seamlessly operate inference workloads using Elotl Nova for multi-cluster orchestration across a fleet of Kubernetes clusters.
What the audience will learn:

• Learn how to plan infra for various components of GenAI stack on Kuberenetes
• Learn how to perform multi-cluster orchestration
• Learn about operational concerns to keep in mind while deploying GenAI at scale

***

Scaling Multi-GPU Multi-Node Deployments with Triton Inference Server - Ryan McCormick (Nvidia)
05:20 PM - 05:40 PM
Description: NVIDIA Triton Inference Server is an open-source inference serving solution that simplifies the production deployment of AI models at scale. With a uniform interface and standard set of metrics, developers can easily deploy models across many different frameworks (TensorRT-LLM, vLLM, ONNX, PyTorch, OpenVINO, and more) on multiple types of hardware (CPU and GPU). Come learn how you can deploy multi-GPU and multi-node models at scale across a cluster of nodes.
What the audience will learn:

• Learn about Nvidia Triton Inference server
• Learn how to deploy Triton on multiple types of hardware
• Learn how to deploy multi-GPU multi-node models at scale across a cluster of nodes

***

Ozone CSI Primer - Prashant Pogde (Cloudera)
05:40 PM - 06:00 PM
Description: This talk covers how Apache Ozone would be an option for read-write-many semantic persistent volume claims in Kubernetes environments. This talk will also provide a demo of how the applications can use Apache Ozone storage option Kubernetes environment via a CSI plugin.
What the audience will learn: The audience will learn about what is CSI and how the CSI option is provided for Apache Ozone in the Kubernetes environment.

***

Ozone Performance and Scale - Ritesh Shukla & Duong Nguyen (Cloudera) 06:00 PM - 06:20 PM
Description: Architectural benefits of Ozone for scale and performance over other S3/HDFS Object Stores. Why architecture matters not just for scale and performance but also performance of day 2 operations that occur in the background. This talk also covers an overview of projects completed, impact on performance and scale and future projects underway.
What the audience will learn:

• Key benefits of Ozone
• Deep dive on architecture and operational model
• Technical deep dive on performance focused projects and future road map