Dynamic Pricing
Last updated
Last updated
In the fast-paced world of cloud computing and AI, the demand for GPU resources fluctuates rapidly. Our new dynamic pricing strategy aims to optimize resource allocation by implementing a quadratic demand model. This approach ensures that prices respond more aggressively to high demand, promoting efficient resource utilization and improved accessibility for all users.
More responsive to high-demand situations
Encourages efficient resource usage during peak times
Provides better price signals to users about resource scarcity
Allows for more nuanced pricing across different demand levels
Our new model uses a quadratic equation to calculate the demand factor, which is then used to adjust the base price of resources in real-time.
Where:
Base Price: The standard price set for the resource
Demand Factor: A multiplier based on current demand (range: 1.0 to 5.0)
The demand factor is now calculated using a quadratic function:
Where:
H: Historical usage factor (0 to 1)
C: Current resource availability factor (0 to 1)
w1 = 0.35 (weight for historical usage)
w2 = 0.65 (weight for current availability)
This quadratic formula ensures that the demand factor grows more rapidly as demand increases, with a maximum value of 5.0.
Before applying our dynamic pricing model, we first establish a base price for each type of computing resource. This base price is derived from the prices set by our various Computing Providers (CPs).
Each CP sets their own prices for different hardware resources, including:
CPU (price per core per hour)
GPU (price per GPU per hour)
Memory (price per GB per hour)
Storage (price per GB per hour)
For each type of resource, we calculate a Weighted Arithmetic Mean (WAM) across the available resource from all CPs. This WAM becomes our base price for that resource.
The formula for WAM is:
Where:
Price_i is a unique price point set by one or more CPs
Weight_i is the number of CPs that have set their price to Price_i
This approach ensures that our base price is more heavily influenced by the most common price points among our CPs.
Grouping by unique prices:
WAM Calculation:
This WAM of $0.0115 would then be used as the base price for memory in our dynamic pricing calculations.
For each hardware configuration we offer, we calculate a composite base price by summing the WAM prices of its components:
We analyze the past 30 days of usage data, focusing on patterns in daily and weekly usage.
Calculation:
This factor considers the current occupation rate of resources, but only starts contributing when more than 40% of resources are occupied.
Calculation:
Historical Usage: Low (H = 0.2)
Current Availability: 45% resources occupied (C = 0.083)
Historical Usage: Average (H = 0.5)
Current Availability: 70% resources occupied (C = 0.5)
Historical Usage: High (H = 0.8)
Current Availability: 90% resources occupied (C = 0.833)
Historical Usage: Very High (H = 1.0)
Current Availability: 100% resources occupied (C = 1.0)
This comprehensive dynamic pricing strategy ensures that our prices reflect both the most common price points among our Computing Providers and the current market demand. By using a frequency-weighted average of CP prices, we establish a fair baseline that represents the consensus pricing in our provider ecosystem. The subsequent application of our quadratic dynamic pricing model then allows us to respond effectively to changes in demand, ensuring efficient resource allocation and maximizing value for all stakeholders in our computing ecosystem. This approach not only captures market trends in resource pricing but also adapts quickly to shifts in user demand, creating a responsive and efficient pricing mechanism.
