Fixed and Market Pricing for Cloud Services
We study a model of congestible resources, where pricing and scheduling are intertwined. Motivated by the problem of pricing cloud instances, we model a cloud computing service as linked $GI/GI/\cdot$ queuing systems where the provider chooses to offer a fixed pricing service, a dynamic market based service, or a hybrid of both, where jobs can be preempted in the market-based service. Users (jobs), who are heterogeneous in both the value they place on service and their cost for waiting, then choose between the services offered. Combining insights from auction theory with queuing theory we are able to characterize user equilibrium behavior, and show its insensitivity to the precise market design mechanism used. We then provide theoretical and simulation based evidence suggesting that a fixed price typically, though not always, generates a higher expected revenue than the hybrid system for the provider.
💡 Research Summary
The paper investigates how a cloud‑service provider should price and schedule its congestible resources when users differ in both the value they place on service and their willingness to wait. The authors model the cloud as a network of linked GI/GI/· queues, each representing a processing stage (e.g., authentication, deployment, execution). Three pricing regimes are considered: a Fixed‑price service where every job pays a single posted price, a Market‑based service where jobs submit bids and higher bidders can preempt lower‑priority jobs, and a Hybrid service that lets users choose between the two.
Users are heterogeneous: each job i has a private value v_i for completion and a waiting‑cost parameter c_i that penalizes delay. Given a pricing regime, a job’s expected utility is U_i = v_i – p_i – c_i·W_i, where p_i is the price paid (fixed or the winning bid) and W_i is the expected waiting time in the corresponding queue. The authors combine auction theory with queueing analysis to derive the equilibrium where every user selects the regime that maximizes U_i. A key insight is that the equilibrium allocation is largely insensitive to the specific auction mechanism (e.g., VCG, Myerson); users’ decisions depend only on their own v_i, c_i, and the average waiting‑time functions of the regimes.
The provider’s problem is to choose the pricing structure that maximizes expected revenue R = Σ_i p_i·q_i, where q_i is the fraction of users who pick regime i. By applying Lagrangian optimization and known results for GI/GI/· queues, the authors obtain closed‑form conditions under which a pure Fixed price yields the highest R. Intuitively, a Fixed price eliminates the overhead of running a real‑time market and avoids the inefficiencies caused by preemptions, which increase average waiting times.
To validate the theory, extensive simulations are performed. Arrival rates λ range from low to high congestion, service times follow both exponential and log‑normal distributions, and v_i and c_i are drawn from normal or exponential distributions to capture a wide spectrum of user heterogeneity. Each scenario is replicated 10,000 times, measuring average revenue, average waiting time, and the share of users selecting each regime. Results confirm that Fixed pricing typically dominates in expected revenue, especially when user values and waiting costs are moderately dispersed. However, when the value distribution is highly skewed—i.e., a small subset of jobs has extremely high v_i—Market or Hybrid schemes can extract additional surplus by charging premium bids, thereby surpassing Fixed pricing.
The paper’s practical implications are clear. For most cloud providers, implementing a simple Fixed‑price offering is cost‑effective and revenue‑optimal, avoiding the engineering complexity of a dynamic auction platform and the performance penalties of preemptive scheduling. Nonetheless, providers serving a niche of premium customers or workloads with very high time‑sensitive value may benefit from a Hybrid or pure Market design that enables price discrimination. The authors also discuss the trade‑off between the added system complexity of preemptive markets and the potential revenue gains, recommending thorough demand analysis before adopting such mechanisms.
Future research directions include extending the model to multi‑resource environments (CPU, memory, bandwidth), incorporating strategic user behavior such as bid shading or job splitting, and designing data‑driven adaptive pricing algorithms that react to real‑time congestion signals. These extensions would bring the theoretical framework closer to the operational realities of modern cloud platforms and enable more nuanced, profit‑maximizing pricing strategies.