Salesforce has successfully implemented a Multi-AZ High Availability (HA) solution for its Agentforce AI foundation using Amazon SageMaker Inference Components (ICs). While ICs significantly reduced infrastructure costs by enabling multiple models to share GPUs, their default placement strategy did not meet Salesforce's compliance requirements for multi-AZ resilience. By leveraging the new `SchedulingConfig` parameter within the `CreateInferenceComponent` API, Salesforce can now control the distribution of IC copies across Availability Zones and instances, ensuring fault tolerance and meeting their internal compliance standards. AI
IMPACT Enables more robust and cost-effective deployment of AI models in production environments by addressing multi-AZ availability concerns.
RANK_REASON This item details the implementation of a specific feature within a cloud provider's ML platform to solve a customer's infrastructure challenge.
Read on AWS Machine Learning Blog →
- Agentforce
- Amazon SageMaker
- AvailabilityZoneBalance
- AWS
- CreateInferenceComponent API
- PlacementStrategy
- SageMaker Inference Components
- Salesforce
- SchedulingConfig
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →