On-Demand Videos
Coupang is a leading e-commerce company in South Korea, with over 50,000 employees and $20+ billion in annual revenue. Coupang's AI platform team builds and manages a large-scale AI platform in AWS for machine learning engineers to train models that enhance and customize product search results and product recommendations for its 100+ million customers.
As the search and recommendation models evolve, optimizing the underlying infrastructure for AI/ML workloads is essential for the e-commerce business. Coupang's platform team actively sought to improve their model training pipeline to boost machine learning engineers' productivity, publish models to production faster, and reduce operational costs.
Coupang focused on addressing several key areas:
- Shortening data preparation and model training time
- Improving GPU utilization in training clusters in different regions
- Reducing S3 API and egress costs incurred from copying large training datasets across regions
- Simplifying the operational complexity of storage system management
In this tech talk, Hyun Jung Baek, Staff Backend Engineer at Coupang, will share best practices for leveraging distributed caching to power search and recommendation model training infrastructure.
Hyun will discuss:
- How Coupang builds a world-class large-scale AI platform for machine learning engineers to deliver better search and recommendation models
- How adding distributed caching to their multi-region AI infrastructure improves GPU utilization, accelerates end-to-end training time, and significantly reduces cross-region data transfer costs.
- How to simplify platform operations and to easily deploy the same architecture to new GPU clusters.
About the Speaker
Hyun Jung Baek is a Staff Backend Engineer at Coupang.
Deepseek’s recent announcement of the Fire-flyer File System (3FS) has sparked excitement across the AI infra community, promising a breakthrough in how machine learning models access and process data.
In this webinar, an expert in distributed systems and AI infrastructure will take you inside Deepseek 3FS, the purpose-built file system for handling large files and high-bandwidth workloads. We’ll break down how 3FS optimizes data access and speeds up AI workloads as well as the design tradeoffs made to maximize throughput for AI workloads.
This webinar you’ll learn about how 3FS works under the hood, including:
✅ The system architecture
✅ Core software components
✅ Read/write flows
✅ Data distribution/placement algorithms
✅ Cluster/node management and disaster recovery
Whether you’re an AI researcher, ML engineer, or infrastructure architect, this deep dive will give you the technical insights you need to determine if 3FS is the right solution for you.
.png)
In this keynote, Calvin Jia will share some of the hottest use cases in Alluxio 2 and discuss the future directions of the project being pioneered by Alluxio and the community. Bin Fan will provide an overview of the growth of Alluxio open-source community with highlights on community-driven collaboration with engineering teams from Microsoft and Alibaba to advance the technology.
Video: Presentation Slides: The Pandemic Changes Everything, the Need for Speed and Resiliency from Alluxio, Inc.
Distributed applications are not new. The first distributed applications were developed over 50 years ago with the arrival of computer networks, such as ARPANET. Since then, developers have leveraged distributed systems to scale out applications and services, including large-scale simulations, web serving, and big data processing. However, until recently, distributed applications have been the exception, rather than the norm. However, this is changing quickly. There are two major trends fueling this transformation: the end of Moore’s Law and the exploding computational demands of new machine learning applications. These trends are leading to a rapidly growing gap between application demands and single-node performance which leaves us with no choice but to distribute these applications. Unfortunately, developing distributed applications is extremely hard, as it requires world-class experts. To make distributed computing easy, we have developed Ray, a framework for building and running general-purpose distributed applications.
We introduce Data Orchestration Hub, a management service that makes it easy to build an analytics or machine learning platform on data sources across regions to unify data lakes. Easy to use wizards connect compute engines, such as Presto or Spark, to data sources across data centers or from a public cloud to a private data center. In this session, you will witness the use of “The Hub” to connect a compute cluster in the cloud with data sources on-premises using Alluxio. This new service allows you to build a hybrid cloud on your own, without the expertise needed to manage or configure Alluxio.
In this keynote, you will learn about the evolution of the global data platform at Rakuten spread across multiple regions, and clouds. In addition, you will hear about the journey across the years, and the use of data orchestration for multiple use cases.
Over the years, Alluxio has grown significantly to be the data orchestration framework for the cloud. The community developers and users have contributed a lot of effort and innovation to make Alluxio the system it is today. There are many users and companies deploying Alluxio at very large scale, and with the large scale, comes different types of challenges.
In this talk, I will introduce the high-level architecture of the current system, and present the various components of Alluxio. Also, I will discuss some of the main challenges of large scale Alluxio deployments, and the lessons we learned from those environments. This talk will detail some of the major scalability improvements added in the past several months, and how users can benefit from the changes.
We are extremely excited to announce the release of Alluxio 2.4.0!
Alluxio 2.4.0 focuses on features critical to large scale, production deployments in Cloud and Hybrid Cloud environments. Features such as highly scalable metadata journaling, aggregate cluster metrics monitoring, and automated detection of JVM pauses further improve Alluxio’s suitability for demanding workloads. Devops tools are also key for triaging issues when they occur. In Alluxio 2.4 we further improve the cluster wide log collection framework. Finally, Alluxio is continually expanding its state of the art integrations with frameworks and storage systems. Alluxio 2.4 introduces and improves integrations with Kubernetes, Azure Data Lake Storage, and Apache Ozone. Alluxio 2.4 is also the first Alluxio release that has support for Java 11.
In this Office Hour, we will go over:
- Expanded metadata service
- Cloud native deployment
- Simplified DevOps and system monitoring
- Support for Java 11
In this talk, we describe the architecture to migrate analytics workloads incrementally to any public cloud (AWS, Google Cloud Platform, or Microsoft Azure) directly on on-prem data without copying the data to cloud storage.
In this Office Hour:
- We will go over an architecture for running elastic compute clusters in the cloud using on-prem HDFS.
- Have a casual online video chat with Alluxio Open Source core maintainers to address any Alluxio related questions from our community members
Over the last few years, organizations have worked towards the separation of storage and compute for a number of benefits in the areas of cost, data duplication and data latency. Cloud resolves most of these issues but comes to the expense of needing a way to query data on remote storages. Alluxio and Presto are a powerful combination to address the compute problem, which is part of the strategy used by Simbiose Ventures to create a product called StorageQuery – A platform to query files in cloud storages with SQL.
This talk will focus on:
- How Alluxio fits StorageQuery’s tech stack;
- Advantages of using Alluxio as a cache layer and its unified filesystem
- Development of new under file system for Backblaze B2 and fine-grained code documentation;
- ShannonDB remote storage mode.
Alluxio 2.3 was just released at the end of June 2020. Calvin and Bin will go over the new features and integrations available and share learnings from the community. Any questions about the release and on-going community feature development are welcome.
In this Office Hour, we will go over:
- Glue Under Database integration
- Under Filesystem mount wizard
- Tiered Storage Enhancements
- Concurrent Metadata Sync
- Delegated Journal Backups
The hybrid cloud model, where cloud resources run Spark or Presto jobs against data stored on-premises, is an appealing solution to reduce resource contention in on-premise environments while also saving in overall costs. One key flaw in a hybrid model is the overhead associated with transferring data between the two environments. Data and metadata locality within the compute application must be achieved in order to maintain the similar performance of analytics jobs as if the entire workload was run on-premises.
In this office hour, we demonstrate how a “zero-copy burst” solution helps to speed up Spark and Presto queries in the public cloud while eliminating the process of manually copying and synchronizing data from the on-premise data lake to cloud storage. This approach allows compute frameworks to decouple from on-premise data sources and scale efficiently by leveraging Alluxio and public cloud resources such as AWS.
We will cover:
- Typical challenges of moving data to the cloud and expanding compute capacity.
- Details about “zero-copy” hybrid cloud solution for burst computing
- A demo of running Presto analytic queries using remote on-prem HDFS data with Alluxio deployed in AWS EMR
As the amount of data analyzed and stored continues to grow exponentially, fixed on-premises infrastructure like Apache Hadoop data lakes becomes costly. Add to that the need to support newer and popular frameworks on an already busy data lake, it is not uncommon to see Hadoop-based data lakes running at beyond 100% utilization and hybrid processing split between physical and cloud infrastructure. As a result, companies are looking to leverage the flexibility and cost savings of the cloud.
Join us for this tech talk where we will show you how Alluxio can help burst your private computing environment to Google Cloud, minimizing costs and I/O overhead. Alluxio coupled with Google’s open source data and analytics processing engine, Dataproc, enables zero-copy burst for faster query performance in the cloud so you can take advantage of resources that are not local to your data, without the need for managing the copying or syncing of that data.
We’ll also show a demo on how to get up and running with Alluxio and Dataproc, including how to:
- Setup your hybrid environment between your private datacenter and Google Cloud Platform
- Burst a Spark based machine learning algorithm to Dataproc while accessing on-prem data
- Scale analytic workloads directly on data on-prem without copying and synchronizing the data into the cloud