Introduction

As AI adoption accelerates across industries, businesses face an undeniable truth - AI is only as powerful as the data that fuels it. Totruly harness AI's potential, organizations must effectively manage, store,andprocess high-scale data while ensuring cost efficiency, resilience, performance and operational agility. 

At CiscoSupport Case Management -IT, we confronted this challenge head-on. Our team deliversacentralized IT platform that manages the entire lifecycle of Cisco product and service cases. Our mission is to provide customers withthe fastest and most effective case resolution, leveraging best-in-class technologies and AI-driven automation. We achieve this while maintaining a platform that is highly scalable, highly available, andcost-efficient. To deliver the best possible customer experience, we must efficiently store and process massive volumes of growing data. This data fuels and trains our AI models, which power critical automation solutions to deliver faster and more accurate resolutions.Our biggest challenge was striking the right balance between building ahighly scalable and reliabledatabasecluster while ensuringcost and operational efficiency. 

Traditional approaches to high availability often rely on separate clusters per datacenter, leading to significant costs, not just for the initial setup but to maintain and manage the data replication process and high availability. However, AI workloads demand real-time data access, rapid processing, and uninterrupted availability, something legacy architectures struggle to deliver. 

So, how do you architect a multi-datacenterinfrastructure that can persist and process massive data to support AI and data-intensive workloads, all while keeping operational costs low? That's exactly the challenge our team set out to solve. 

In this blog, we'll explore how we built an intelligent, scalable, and AI-ready data infrastructure that enables real-time decision-making, optimizes resourceutilization, reduces costs and redefines operational efficiency. 

Rethinking AI-ready case management at scale

In today's AI-driven world, customer support is no longer just about resolving cases, it's about continuously learning and automating to make resolution faster and better while efficiently handling the cost and operational agility.  

The same rich dataset that powers case management must also fuel AI models and automation workflows, reducing case resolution time from hours or days to mere minutes, which helps in increased customer satisfaction. 

This created a fundamental challenge:decoupling the primary databasethat serves mainstream case management transactional system from an AI-ready, search-friendly database, a necessity for scaling automation without overburdening the core platform. While the idea made perfect sense, it introduced two major concerns: cost and scalability. As AI workloads grow, so does the amount of data. Managing this ever-expanding dataset while ensuring high performance, resilience, and minimal manual intervention during outages required an entirely new approach. 

Rather than following the traditional model of deploying separate databaseclusters per data center for high availability, we took a bold step toward building asingle stretched database cluster spanning multiple data centers. This architecture not only optimized resource utilization and reduced both initial and maintenance costs but also ensured seamless data availability. 

By rethinking traditional indexdatabase infrastructure models, we redefined AI-powered automation for Cisco case management, paving the way for faster, smarter, and more cost-effective support solutions. 

How we solved it -The technology foundation

Building a multi-data center modernindex database cluster required a robust technological foundation, capable of handlinghigh-scale data processing,ultra-low latency for faster data replication, and careful design approach to build a fault-tolerance to support high availabilitywithout compromisingperformance, or cost-efficiency. 

Network Requirements

A key challenge in stretching anindex database cluster across multiple data centersisnetwork performance. Traditionalhigh availability architectures rely onseparate clusters per data center, often struggling withdata replication, latency, and synchronization bottlenecks. To begin with,we conducted adetailed network assessment across our Cisco data centers focusing on: 

• Latency and bandwidth requirements-Our AI-powered automation workloads demand real-time data access. We analyzed latency and bandwidth between two separate data centers to determine if a stretched cluster was viable.  
• Capacity planning-We assessed our expected data growth, AI query patterns, and indexing rates to ensure that the infrastructure could scale efficiently. 
• Resiliency and failover readiness-The network needed to handle automated failovers, ensuring uninterrupted data availability, even during outages. 

How Cisco'shigh-performance data center paved the way

Cisco's high-performance data center networking laida strongfoundation for building the multi-data center stretch single database cluster.The latency and bandwidth provided by Cisco data centers exceededour expectation to confidently moveon to the next step of designing a stretch cluster. Our implementationleveraged:

• Cisco ApplicationCentric Infrastructure (ACI)-Offered a policy-driven, software-defined network, ensuring optimized routing, low-latency communication, and workload-aware traffic management between data centers.  
• Cisco Application Policy Infrastructure Controller (APIC) and Nexus 9000 Switches-Enabled high-throughput, resilient, and dynamically scalable interconnectivity, crucial for quick data synchronization across data centers. 

The Cisco data center and networking technology made this possible. It empowered Cisco IT to take this idea forward and enabled us to build this successful cluster which saves significant costs and provides high operational efficiency.

Our implementation -The multi-data center stretch cluster leveraging Cisco datacenterand network power

With the right network infrastructure in place, we set out to build a highly available, scalable, and AI-optimized database cluster spanning multiple data centers.

Cisco multi-data center stretch Index database cluster

Key design decisions

• Single logical, multi-data center cluster for real-time AI-driven automation-Instead of maintaining separate clusters per data center which doubles costs, increases maintenance efforts, and demands significant manual intervention, we built a stretched cluster across multiple data centers. This design leverages Cisco's exceptionally powerful data center network, enabling seamless data synchronization and supporting real-time AI-driven automation with improved efficiency and scalability.  
• Intelligent data placement and synchronization -We strategically position data nodes across multiple data centers using custom data allocation policies to ensure each data center maintains a unique copy of the data, enhancing high availability and fault tolerance. Additionally, locally attached storage disks on virtual machines enable faster data synchronization, leveraging Cisco's robust data center capabilities to achieve minimal latency. This approach optimizes both performance and cost-efficiency while ensuring data resilience for AI models and critical workloads. This approach helps in faster AI-driven queries, reducing data retrieval latencies for automation workflows. 
• Automated failover and high availability -With a single cluster stretched across multiple data centers, failover occurs automatically due to the cluster's inherent fault tolerance. In the event of virtual machine, node, or data center outages, traffic isseamlesslyrerouted to available nodes or data centerswith minimal to no human intervention. This is made possible by therobust network capabilitiesof Cisco's data centers, enabling data synchronization inless than 5 millisecondsfor minimal disruption and maximum uptime. 

Results

By implementing these AI-focused optimizations, we ensured that the case management system could power automation at scale, reduce resolution time, and maintain resilience and efficiency. The results were realized quickly.

• Faster case resolution:Reduced resolution time from hours/days to just minutes by enabling real-time AI-powered automation. 
• Cost savings: Eliminatedredundant clusters, cutting infrastructure costs while improvingresource utilization.  
  • Infrastructure cost reduction: 50% savings per quarter by limiting it to one single-stretch cluster, by completing eliminating a separate backup cluster. 
  • License cost reduction: 50% savings per quarter as the licensing is required just for one cluster. 
• Seamless AI model training and automation workflows:Provided scalable, high-performance indexing for continuous AI learning and automation improvements. 
• High resilience and minimal downtime:Automated failovers ensured 99.99% availability, even during maintenance or network disruptions. 
• Future-ready scalability:Designed to handle growing AI workloads, ensuring that as data scales, the infrastructure remains efficient and cost-effective.

By rethinking traditional high availability strategies and leveraging Cisco's cutting-edge data center technology, we created a next-gen case management platform-one that's smarter, faster, and AI-driven.

Additional resources:

• Cisco on Cisco