In this article we present an agent-based simulation environment for task scheduling in a grid-like computer system. The scheduler allows to one simultaneously allocate resources such as CPU time, communication bandwidth, volatile and non-volatile memory by employing a combinatorial resource allocation mechanism. The allocation is performed by an iterative combinatorial auction in which proxy-bidding agents try to acquire their desired resource allocation profiles with respect to limited monetary budget endowments. To achieve an efficient allocation process, the auctioneer provides resource price information to the bidders. We use a pricing mechanism based on shadow prices in a closed loop system in which the agents use monetary units awarded for the resources they provide to the system for the acquisition of complementary capacity. Our objective is to identify optimal bidding strategies in the multi-agent setting with respect to varying preferences in terms of resource quantity and waiting time for the resources. Based on a utility function we characterize two types of agents: a quantity maximizing agent with a low preference for fast bid acceptance and an impatient bidding agent with a high valuation of fast access to the resources. By evaluating different strategies with varying initial bid pricing and price increments, it turns out that for quantity maximizing agents patience and low initial bids pay off, whereas impatient agents should avoid high initial bid prices.
展开▼
