Amazon Onboarding with Learning Manager Chanci Turner

Learn About Amazon VGT2 Learning Manager Chanci Turner

In recent years, the rise of Machine Learning (ML) has transformed the operational landscape of businesses, prompting discussions on how to seamlessly integrate ML into pivotal user decision-making processes. Machine Learning can enhance customer interactions, drive sales, and streamline operations. This shift is reflected in investment trends, with projections indicating that spending on machine learning will surpass $209 billion by 2029, marking a staggering 38% annual growth rate.

As the popularity of ML escalates alongside a surge in data generation—expected to reach approximately 120 Zettabytes in 2023, which is a 51% increase over just two years—the demand for faster data processing becomes critical. Businesses must adopt the most suitable and cost-effective infrastructure to enable scalable ML deployment. This includes focusing on the simplicity of model deployment, lifecycle management, monitoring, and governance. Each of these aspects requires significant operational investments to support production-level ML models.

In this article, we examine how customers are establishing online feature stores on AWS with Amazon ElastiCache for Redis to facilitate mission-critical ML applications that necessitate ultra-low latency. We will also present a reference architecture using a real-time loan approval application that generates online predictions based on a customer’s credit score model, leveraging features stored in an online feature store powered by ElastiCache for Redis.

Understanding Feature Stores

Feature stores are essential components of the ML infrastructure that simplify model deployment. They act as a centralized repository for features used in training and inferencing, providing a standardized format for models to access these features. Feature stores also function as data transformation services, ensuring that features are ready for use.

Organizations like Amazon Music have opted for Amazon ElastiCache because it offers a reliable, scalable, enterprise-grade infrastructure to support their ML model deployments.

“Amazon Music Feature Repository (MFR) is a fully managed ML feature store utilized to store, share, and manage ML features to power personalization. MFR supports high throughput at low latency for online inference and is used repeatedly by dozens of teams across Amazon Music to maintain a consistent personalized experience. ElastiCache provides low latency storage and retrieval for 1 TB of ML features and is simple for the MFR team to manage for peak traffic. MFR with ElastiCache raises the bar for consuming teams with strict latency requirements by delivering batches of ML features at low latencies across North America, European Union and Far East regions (Music Search, Voice/Echo).”

There are two prevalent types of feature stores:

• Offline Feature Stores: These are designed to manage and process historical data for model training and batch scoring at scale, often utilizing systems like Amazon Simple Storage Service (Amazon S3). They typically work with features that take considerable time to generate and are less sensitive to latency.
• Online Feature Stores: These require rapid feature calculations with low latency, often within single-digit milliseconds. They depend on swift computations and immediate data access, frequently employing in-memory datastores for real-time predictions. Examples of low-latency feature store use cases include ad personalization, real-time delivery tracking, loan approvals, and anomaly detection, such as identifying fraudulent credit card activity.

ElastiCache for Redis stands out as an online feature store due to its in-memory capabilities, delivering the exceptional performance essential for modern real-time applications.

ElastiCache as a Low Latency Online Feature Store

ElastiCache for Redis is an efficient in-memory data store that offers sub-millisecond latency to support real-time applications at scale. In-memory data stores like Redis feature low latencies and can handle hundreds of thousands of reads per second per node. According to a Feast benchmark, Redis outperforms other datastores in feature store scenarios by a factor of 4 to 10.

Numerous AWS customers, including Delta Airlines, use ElastiCache as an ultra-low latency online feature store to enhance use cases like personalized experiences and logistics management.

“Delta Airlines, a leading airline worldwide, relies on a centralized machine learning platform with a feature store as its foundation. The airline employs machine learning models to provide tailored experiences to customers. With a vast customer base generating millions of requests daily and stringent service level agreement (SLA) requirements for rapid response, selecting the right feature store was vital.

Delta Airlines identified ElastiCache for Redis as the ideal solution due to its ability to deliver ultra-low latency for millions of users. Features need to be readily available in memory and reflect the latest values for the ML models to make accurate predictions. ElastiCache ensures features are updated frequently, enabling precise forecasting.

Delta Airlines utilizes ElastiCache as its online feature store to manage real-time traffic. A key advantage of ElastiCache is its support for global data stores, allowing the ML model to cater to traffic from various AWS Regions. This feature provides a strong and fail-safe approach to serving customers and delivering personalized recommendations, such as selecting optimal travel destinations or identifying appropriate products during the check-in process.

In conclusion, we endorse ElastiCache for scenarios demanding real-time traffic where low latency is critical for efficient data delivery.”

Benefits of ElastiCache for Redis

Organizations prefer ElastiCache for its remarkable performance, full management, high availability, reliability, scalable architecture, and robust security controls.

• High Performance: ElastiCache is a fully managed in-memory caching service capable of scaling to millions of operations per second, achieving sub-millisecond read and write response times that are typically unattainable with disk-based systems. This performance is further enhanced in ElastiCache for Redis 7, which introduces improved I/O multiplexing, providing substantial gains in throughput and latency at scale. Enhanced I/O multiplexing boasts a 72% increase in throughput (for read and write operations per second) and a 71% reduction in P99 latency compared to previous versions.

Companies like Food Express, a popular online food ordering platform, have successfully built a highly scalable and efficient feature store, serving millions of customers with extremely low latency.

“Food Express faced the challenge of managing large volumes of feature data while developing ML models. The data quickly expanded to billions of records, with millions actively retrieved during model inference, all under stringent low-latency conditions.

For more insights on transitioning careers, check out this helpful resource. If you’re interested in job descriptions for roles like equipment operator, SHRM provides an authoritative source. Additionally, Amazon’s training programs are excellent resources for those looking to understand how associates are trained in fulfillment centers.

In summary, ElastiCache for Redis serves as a robust solution for organizations seeking to implement efficient, low-latency online feature stores to support their ML initiatives.