- مبلغ: ۸۶,۰۰۰ تومان
- مبلغ: ۹۱,۰۰۰ تومان
Traditionally, heavy computational tasks were performed on a dedicated infrastructure requiring a heavy initial investment, such as a supercomputer or a data center. Grid computing relaxed the assumptions of the fixed infrastructure, allowing the sharing of remote computational resources. Cloud computing brought these ideas into the commercial realm and allows users to request on demand an essentially unlimited amount of computing power. However, in contrast to previous assumptions, this computing power is metered and billed on an hour-by-hour basis. In this paper, we are considering applications where the output quality increases with the deployed computational power, a large class including applications ranging from weather prediction to financial modeling. We are proposing a computation scheduling that considers both the financial cost of the computation and the predicted financial benefit of the output, that is, its value of information (VoI). We model the proposed approach for an example of analyzing real-estate investment opportunities in a competitive environment. We show that by using the VoI-based scheduling algorithm, we can outperform minimalistic computing approaches, large but fixedly allocated data centers and cloud computing approaches that do not consider the VoI.
The amount of high performance computing performed at a research institution or business used to be limited by the available computational facilities. The decision to invest in such facilities were justified and made years in advance. The advent of cloud computing made the decision to perform high performance computation on thousands of computer cores a decision that can be taken on a minute’s notice. In this paper, we argued that the ability to make this decision very quickly does not reduce the need to analyze whether the computational expenses are justified or not. We discussed that many modern high performance computing applications are elastic in term of computational power - additional computation improves the quality of results, but often with a curve of diminishing returns. We argue that a convenient technique to create efficient decision making approaches is to use the concept of “value of information” - to try to quantify the amount of financial benefit a certain calculation can gain us, and use this value when making scheduling decisions.
We illustrate the proposed model with the example of an investor who is analyzing real estate investment opportunities. We compare approaches that assume minimal, no-cost computational analysis with the cost of maintaining a private data center, buying computational power on the cloud and, finally, with a VoI-informed, cloud-based scheduling approach. We find that the VoI approach clearly outperforms every other approach accross a wide range of opportunity arrival rates and competition intensity values.