Dynamic Capacity Planning in Data Centers for Latency Sensitive Workloads like Immersive Media & AR/VR: A Decision-Engine Approach
Author(s):
Anurag Reddy*, Anil Naik and Sandeep Reddy
This paper explores the intricacies of data center planning, specifically focusing on the management of internet traffic and the optimization of capacity for edge servers. In an era where the internet is integral to daily life, effective capacity planning is essential to meet the escalating demands on network infrastructure. The paper introduces factors influencing internet and traffic demands, highlights the concept of capacity planning, and outlines key elements such as traffic analysis, scalability, edge server deployment, and resource monitoring, with a special focus on the implications in futuristic applications like Augmented & Virtual Reality Two primary methodologies for ensuring required capacity are discussed: the Built to Forecast Methodology and the Decision Engine Methodology. The Built to Forecast approach relies on historical data and usage patterns for proactive infrastructure building, while the Decision Engine approach utilizes intelligent algorithms for dynamic capacity adjustments based on real-time data specific to AR/VR workloads. The latter incorporates push-pull mechanisms and multi-echelon inventory management, addressing the unique challenges posed by the dynamic nature of AR/VR traffic. The paper delves into challenges and inefficiencies associated with the Built to Forecast Methodology, emphasizing scalability issues, the risk of technological obsolescence, inefficient energy consumption, and operational strain. In response, the Decision Engine Methodology is presented as a more adaptive alternative, incorporating push-pull mechanisms and multi-echelon inventory management for efficient supply chain optimization. The conclusion underscores the critical decision between a recommendation-based model and a built-to-forecast approach, emphasizing the trade-offs involved. The optimal choice depends on organizational requirements, risk tolerance, and industry characteristics, with the recommendation-based model excelling in scenarios of high demand variability. Ultimately, the paper advocates for a tailored and effective inventory management strategy aligned with the unique needs and circumstances of the organization.
