Important: This Guidance requires the use of AWS CodeCommit, which is no longer available to new customers. Existing customers of AWS CodeCommit can continue using and deploying this Guidance as normal.
This Guidance helps AdTech users build, train, and deploy machine learning models to an ad auction server application. This helps reduce unanswered bid requests by decoupling model creation and model consumption into separate environments to enable independent scale and security measures.
Architecture Diagram
Step 1
Copy raw OpenRTB log data into an Amazon Simple Storage Service (Amazon S3) bucket.
Step 2
Amazon SageMaker Studio deploys an Amazon EMR cluster to process raw OpenRTB data.
Step 3
Amazon SageMaker Pipelines is triggered by a user to launch the pre-processing and training steps.
Step 4
In the pre-processing step, the raw data is transformed in Amazon EMR and stored in another Amazon S3 bucket, with the transformation steps defined in a SageMaker preprocessing notebook.
Step 5
A machine learning (ML) training job uses the pre-processed data, and outputs trained models to the SageMaker model registry.
Step 6
Register the last version, save the trained model to an Amazon S3 bucket, and register metadata in AWS AppConfig.
Step 7
In a separate AWS environment, the ad auction server and the bid filtering applications are deployed to Amazon Elastic Container Service (Amazon ECS) as individual containers.
Step 8
The bid filtering application pulls the latest version from the model registry and then downloads the model from Amazon S3.
Step 9
The publisher issues auction requests to the ad auction server.
Step 10
The auction server calls the bid filter. The bid filter predicts demand side platform’s (DSPs) likelihood of bidding.
Step 11
Bids to DSP below the threshold are filtered, eliminating the associated data transfer cost.
Step 1
Copy raw OpenRTB log data into an Amazon Simple Storage Service (Amazon S3) bucket.
Step 2
Amazon SageMaker Studio deploys an Amazon EMR cluster to process raw OpenRTB data.
Step 3
Amazon SageMaker Pipelines is triggered by a user to launch the pre-processing and training steps.
Step 4
In the pre-processing step, the raw data is transformed in Amazon EMR and stored in another Amazon S3 bucket, with the transformation steps defined in a SageMaker preprocessing notebook.
Step 5
A machine learning (ML) training job uses the pre-processed data, and outputs trained models to the SageMaker model registry.
Step 6
Register the last version, save the trained model to an Amazon S3 bucket, and register metadata in AWS AppConfig.
Step 7
In a separate AWS environment, the ad auction server and the bid filtering applications are deployed to Amazon Elastic Container Service (Amazon ECS) as individual containers.
Step 8
The bid filtering application pulls the latest version from the model registry and then downloads the model from Amazon S3.
Step 9
The publisher issues auction requests to the ad auction server.
Step 10
The auction server calls the bid filter. The bid filter predicts demand side platform’s (DSPs) likelihood of bidding.
Step 11
Bids to DSP below the threshold are filtered, eliminating the associated data transfer cost.
Well-Architected Pillars
The AWS Well-Architected Framework helps you understand the pros and cons of the decisions you make when building systems in the cloud. The six pillars of the Framework allow you to learn architectural best practices for designing and operating reliable, secure, efficient, cost-effective, and sustainable systems. Using the AWS Well-Architected Tool, available at no charge in the AWS Management Console, you can review your workloads against these best practices by answering a set of questions for each pillar.
The architecture diagram above is an example of a Solution created with Well-Architected best practices in mind. To be fully Well-Architected, you should follow as many Well-Architected best practices as possible.
-
Operational ExcellenceRead the Operational Excellence whitepaper
This Guidance is represented by two AWS Cloud Development Kit (AWS CDK) stacks. You can deploy changes to the application and infrastructure by applying the best practices from AWS CDKs. Programmatic access key, single sign-on, or federation are some of the authentication methods used in the AWS CDK CLI.
An Amazon CloudWatch dashboard provides business metrics for monitoring. You can configure CloudWatch alarms to meet your operational needs.
-
SecurityRead the Security whitepaper
To improve privacy and security, Amazon Virtual Private Cloud (Amazon VPC) service endpoints are used. Managed services such as AWS Lambda and Amazon ECS are used to reduce the security maintenance tasks. Amazon S3 buckets used in this Guidance are encrypted and blocked from public access. Amazon Elastic Block Store (Amazon EBS) volumes are encrypted using the customer managed AWS Key Management Service (AWS KMS) key. The source code for the notebooks is stored in an encrypted AWS CodeCommit repository, and the container images are stored in Amazon Elastic Container Registry (Amazon ECR).
-
ReliabilityRead the Reliability whitepaper
This solution uses managed services Amazon EMR, SageMaker, and Amazon ECS to minimize the operational efforts for the solutions. Amazon EMR is designed to handle large-scale data processing. The model training is done on SageMaker that provides the configuration parameters to size the infrastructure according to the demand. The data processing and model training parts of this solution are run periodically in batches. The inference part must be near real-time and run as containers in Amazon ECS, allowing for multi-Availability Zone deployment and auto-scaling to enable high availability.
-
Performance EfficiencyRead the Performance Efficiency whitepaper
For a hands-on experience, you can deploy the provided code in your account from an AWS Region of your choice. You would first use the sample data to start the data processing and machine learning process and then use a trained model to make inferences. To start tailoring this Guidance to your needs, you can incorporate your own data, modify the machine learning model training, and possibly adjust the inference part.
-
Cost OptimizationRead the Cost Optimization whitepaper
This Guidance is designed to support managed services that can run in any AWS region without incurring additional licensing costs. Deploy to the AWS Region that best fits your needs. The components are managed individually to incur cost only when used. Data transformation uses Amazon EMR clusters to provide the best cost to performance ratio when processing large data sets. From SageMaker studio, clusters can be created and ended to manage costs. The inference component uses container-based deployment to match resource consumption with actual demand.
-
SustainabilityRead the Sustainability whitepaper
This Guidance focuses on shared storage and single sources of truth to avoid data duplication and reduce the total storage requirements of your workload. Only fetch data from shared storage if necessary. Detach unused volumes in order to make more resources available. You can change the utilized compute resources based on the actual need. Adjust provisioned resources based on the actual demand. To understand utilization and the right-size of deployed resources, you can use CloudWatch metrics.
Implementation Resources
The sample code is a starting point. It is industry validated, prescriptive but not definitive, and a peek under the hood to help you begin.
Related Content
Disclaimer
The sample code; software libraries; command line tools; proofs of concept; templates; or other related technology (including any of the foregoing that are provided by our personnel) is provided to you as AWS Content under the AWS Customer Agreement, or the relevant written agreement between you and AWS (whichever applies). You should not use this AWS Content in your production accounts, or on production or other critical data. You are responsible for testing, securing, and optimizing the AWS Content, such as sample code, as appropriate for production grade use based on your specific quality control practices and standards. Deploying AWS Content may incur AWS charges for creating or using AWS chargeable resources, such as running Amazon EC2 instances or using Amazon S3 storage.