Auction allocation of computing resources

Abstract

tandard methods for allocadng computing resources normally employ schedulers and either queue or priority schemes. Alternative methods utilizing marketlikc processes are being investigated, with direct applicability to evolving distributed systems. In this article, we present results of simulations of an auction allocation in which computing tasks are provided sufficient intclligcMUf to acquire resources by offering, bidding, and exchanging them {or funds.There are a number of important consequences of our experiments. First, we confirmed that a resource valuation variable in our model fiiin tions as a price, and is correlated to the arrival rates and complexities of the tasks. This price variable complements other measures, such as processing time, overhead costs, resource utilization, and efficiency, that assist in monitoring overall system performance. Second, such research provides a vicarious opportunity to explore potential non-von Neumann approaches to computer architectures and operating systems. Third, we note that these models are self-regulating, manifest bidding that increases with capacity, and successfully ration resources. Last, prior to their fully developed implementations, internal decentralized resource allocation clearly indicates how either external costing procedures or internal allocation in the operational phases of the lifecycle can be shifted leftward, or earlier, in the sy.stems development phase.During the past decade, there has been an enormous increase in the availability of computing resources, their connectivity through networks, and their complexity as seen in the variety of emerging parallel and distributed systems. Previously, standalone mainframes processed jobs (tasks) in batch mode, which economized on computing resources while wasting human resources. Time-sharing provided some improvement in system response, but its advantage was overcome by unmet demand, rising costs, and even technical difficulties such as thrashing. As personal computers (PCs) became available, they attracted small, independent tasks away from mainframes.However, complex tasks, particularly those requiring large amounts of numerical computation or extensive database access, usually exceeded the limits of these smaller machines. As PCs began to approach the size and power of workstations, many of these kinds of machines were then linked together in networks, which now demand more flexible and responsive access and allocation.In order to suggest changes in the allocation of computing resources and analyze the effects of their implementation, it is necessary to understand how allocation is characterized. Traditionally, computing resource allocation is accomplished cither internally, through scheduling in operating systems, or externally, through administrative policy. The general mode of external allocation in computing centers uses a combination of priority systems and pricing techniques to reduce utilization, thereby curtailing congestion [16].Other external approaches have attempted to utilize cost-shari...

Combinatorial Auction-Based Mechanisms for VM Provisioning and Allocation in Clouds

“…Gagliano [11] also investigated allocating computing resources through auctions, where the tasks themselves are provided enough intelligence to calculate the bid that is necessary to get the required resources. Recently, researchers investigated economic models for resource allocation in computational grids.…”

Dynamic Auctions for On-Demand Services

“…Finally, this paper is related to another strand in the literature on load balancing and scheduling for computer-based services (see [2] and [15]). …”

Developing a Near-optimal Lowest-consumption Tunnel Lighting System Using Software Agents through Power Line Communications

“…Computational markets are based on computational agents . Computational markets have been suggested as a solution to complex resource allocation, scheduling, and optimization problems (Kurose and Simha, 1989; Lenting and Braspenning, 1993;Gagliano et al, 1995). sketch how direct load management can be modeled as a computational market.…”