Federated Optimizing for Distributing Resources Across Tenants
Prerna Dusi, Md Afzal · 2025
A Federated Optimization Approach for Resource Allocation in Multi-tenant Environments Multitenant applications that provide common application/data/computation resource units for multiple independent users (or organizations) have implemented federated optimization techniques to optimize their utilization. PRIVACY AND AUTONOMY: Multi-tenant systems are designed to provide services to multiple tenants; thus, their privacy and autonomy must be guaranteed. Appropriately then federated optimization provides a broad basis for optimizing resource allocation across multiple tenants without centralizing the data which is an important requirement in situations where data privacy and safety of new data is paramount. In a nut shell, federated optimization for multi-tenant systems is aimed of course at decoupling the local computation for all tenants (each on their own data) from the global knowledge propagation by aggregating the local knowledge after computation of the global model that concerns all tenants. Within this collaborative learning setting, the system can improve its resource allocation plans by learning from each tenant requirements and preferences while ensuring privacy. Resource allocation in these types of systems is a fundamentally hard problem because of contention from multiple tenants that can have distinct priorities, cost objectives, and output requirements. Also, different tenants can have different levels of access to system resources and run under different workloads, leading to increased optimization complexity. Federated optimization algorithms(e.g., federated gradient descent or federated averaging) can be employed to enable resource allocation across tenant to be optimized simultaneously to address this. I would be glad to assist you with only paraphrasing the previous sentence into a human-write-style output. In fact, federated optimization allows for such adaptations in systems to dynamic changes in per-tenant resource requirements and workloads characteristics, by iteratively aggregating the model parameters computed at local sites and updating a shared global model. One of the main focus areas of the federated optimization is multitenant systems as, it can help avoid data confidentiality issues. Since tenants' private data was never stored outside their local environments, the system could provide tailored solutions without revealing sensitive data. That is particularly valid in these areas, where they will need to comply with privacy regulations and compliance requirements, as ones you have in health care, financial and telecommunications. In addition, federated optimization enhances resource allocation in large multi-tenant scenarios with global transfer and computation being vague or impractical.