Overview
Arietta AI, a leader in Applied AI solutions for the pharmaceutical industry, faced significant performance and cost challenges with their x86-based cloud infrastructure. Partnering with the cloud services team at skyPurple Cloud, Arietta AI embarked on a transformative journey during 2023 to migrate key workloads to AWS Graviton processors, resulting in substantial cost savings and performance improvements across their AWS infrastructure.
Initial Objectives
- Reduce cloud expenditure without impacting service quality or availability
- Enhance performance of data-intensive AI applications
- Improve energy efficiency of cloud operations
Operational Challenges
Arietta AI's existing x86-based infrastructure led to several operational challenges:
- High compute costs for database operations, particularly in RDS instances
- Suboptimal performance for AI and machine learning workloads
- Energy inefficiencies contributing to higher operational costs and environmental impact
- Scalability limitations for their growing SaaS offerings
Strategies and Solutions Employed
Database Migration to Graviton
- • Migrate all RDS (Relational Database Service) deployments from x86 processor architecture to AWS Graviton processors
- • Move Production RDS clusters from x86 8xlarge instances to 4xlarge Graviton instances, aiming to halve the cost profile while maintaining performance
- • Comprehensive testing to ensure database compatibility and performance optimization
EC2 & ECS Optimization
- • Transition compatible EC2 instances and ECS tasks to Graviton-based compute resources
- • Optimize container images and application code for ARM architecture where necessary
- • Implement gradual rollout strategy to minimize risk and ensure stability
Performance Tuning
- • Fine-tune database configurations and query optimizations to use Graviton's architecture
- • Adjust machine learning pipelines to take advantage of Graviton's performance characteristics
- • Benchmark and validate performance improvements across all workloads
Monitoring and Analytics
- • Implement comprehensive monitoring to track performance metrics and cost savings throughout the migration
- • Utilize Amazon CloudWatch and AWS Cost Explorer to provide detailed insights into the impact of the Graviton migration
- • Establish baseline metrics and ongoing reporting framework
Results
Financial Impact
- Achieved a 50% reduction in compute costs for RDS instances migrated to Graviton
- Contributed significantly to an overall 43% reduction in AWS spend
Performance Improvements
- Enhanced database performance and reduced latency by adopting AWS Graviton processors
- Increased efficiency in machine learning operations, with some workloads seeing up to 20% improvement in processing times
Operational Benefits
- Improved energy efficiency, with an estimated 20% reduction in power consumption for migrated workloads
- Enhanced scalability, allowing Arietta AI to handle increased workloads without proportional cost increases
Future Steps
- Expand Graviton adoption to additional services and workloads
- Continuously optimize application code and configurations to further use Graviton's capabilities
- Explore Graviton-powered instances for new product developments and AI model training
- Develop a comprehensive strategy to quantify and report on the environmental impact reductions achieved through the Graviton migration
Conclusion
Arietta AI's partnership with the skyPurple Cloud team for the Graviton migration project transformed their cloud infrastructure into a more cost-effective and high-performing environment. The migration to Graviton processors not only yielded significant financial savings but also positioned Arietta AI to sustainably manage their cloud resources moving forward.
The performance improvements aligned well with their operational goals, allowing for more efficient AI and machine learning operations. This case study exemplifies how a targeted migration to AWS Graviton processors can result in substantial financial savings, performance improvements, and increased energy efficiency—serving as a model for similar organizations aiming to optimize their cloud infrastructure while maintaining or enhancing service quality.