By Aaron K. Baughman | Last updated October 12, 2018
Artificial IntelligenceObject StorageCloud
IBM Watson artificial intelligence (AI) has your back when you cannot sleep at night thinking about your ESPN Fantasy Football lineup selections. #FantasyFootballFace. ESPN Fantasy Football with Watson has been designed to perpetually read, understand, and comprehend multimedia and text data. Our fantasy football insights constantly update fantasy football player insights in near real-time with the goal of no downtime. IBM Watson is always on, engaged in fantasy football analysis 24/7 to make you a better player, even when you rest.
The AI system architecture follows a continuous availability (CA) architecture to achieve at least 5 9s of uptime or only 5 minutes of planned and unplanned outages per year. The back-end components that generate the fantasy football predictions, evidence, and trends are run in parallel over three different Out of Region (OoR) centers on the IBM Cloud. To avoid any outage within a particular region, the centers are spread across the world in Dallas, Germany, and the United Kingdom. Each OoR cloud shares Watson components such as Watson Discovery, Cloud Object Storage, and DB2 on Cloud that have their own availability Service Level Agreements (SLAs). DB2 on Cloud runs on a High Availability Disaster Recovery (HADR) architecture to prevent any loss of data. IBM Cloud Object Storage follows a traditional three active site for Continuous Availability (CA) while Watson Discovery distributes its traffic across multiple servers for CA.
Overall, client traffic from mobile devices, desktops, or tablets are sent to the IBM Cloud edge servers. Hundreds of edge servers are combined into a Content Delivery Network (CDN) to accelerate Fantasy Football insight delivery. The CDN protects the continuous availability artificial intelligence insights. Word-2-Vector, deep learning, polynomial regression, decision trees, and simulation algorithms run on a schedule and as players are updated. The sustained volume of millions of users and trillions of insight requests do not have an impact on the AI availability or performance.
Each of the parallel CA artificial intelligence systems are stateless and independent. The output of each system is written into a separate Object Storage as origin. User clients access the origin with the most recent data. The availability of the AI OoR parallel systems is calculated by the product of each component availability as shown in the following image.
All of the Python and the Node.js application availability is determined over time by the IBM Cloud Availability Monitoring service. However, the Python applications use many different Watson services in serial. Each of the Watson services has an availability SLA that can be multiplied together as they are serially used within the Python machine learning pipeline. The following equation shows how to determine serial availability of the AI pipeline.
Even though the Node.js application depends on Object Storage and the IBM Cloud CDN, each call is asynchronous. As such, the parallel availability can be used to determine the Node.js component availability. The nesting of the parallel and serial availability equations provide a way to optimize the performance and availability of the ESPN Fantasy Football with Watson system.
Even when you sleep, you are playing ESPN Fantasy Football smarter with Watson. #WinWithWatson
Check back next time as I demystify deep learning with evidence.
The ESPN Fantasy Football logo is a trademark of ESPN, Inc. Used with permission of ESPN, Inc.
Back to top