An Optimal Virtual Valuation-Based Combinatorial Auction Mechanism for Time-Varying Resource Allocation in Heterogeneous Cloud Services

Jixian Zhang

Abstract:

The resource allocation problem that is posed by
cloud services has long been a popular research topic. The existing
auction mechanisms focus primarily on maximizing social
welfare, but they often result in lower revenue for cloud service
providers. The virtual valuation-based combinatorial auction
(VVCA) mechanism can increase the revenue that is obtained
by service providers while satisfying dominant strategy incentive
compatibility (DSIC). In this study, we innovatively apply the
VVCA mechanism to address a time-varying resource allocation
problem that involves heterogeneous servers (HTs) in cloud services
and effectively increase the revenue that is received by cloud
service providers. We begin by transforming the HT problem into
an integer programming model with time-varying and resource
constraint features. Afterward, we provide the theoretical basis
for using the VVCA mechanism to solve the aforementioned
problem and provide the DSIC proof. On this basis, we design
three progressively more effective mechanisms using the VVCA
mechanism. (1) We develop a random mechanism HT VVCAm
and prove that it has a logarithmic approximation ratio, thus
offering a better lower bound guarantee than the existing
approach does. (2) We propose a gradient-based optimization
mechanism HT VVCA∗ to approximate the optimal revenue. (3)
We design an optimal revenue algorithm called HT VVCANET
on the basis of the transformer architecture that is used in
deep learning; this algorithm achieves a good balance between
execution efficiency and effectiveness. In the experiments, we
implement these mechanisms, which significantly increase the
revenue that is received by cloud service providers over that
yielded by other benchmark mechanisms.