Just in time clouds: uma abordagem baseada em recursos terceirizados para a ampliação da elasticidade de provedores de computação na nuvem.

Abstract

The throughput obtained when executing HTC (High Throughput Computing) applications on a computing infrastructure depends directly on the scale that it offers. In this context, the size of the processing pool is the principal promoter of performance, while the coordination effort involved is the main limiting factor. The paradigm of cloud computing enables the delivery of Information Technology infrastructure in the form of a service that customers purchase on-demand and pay only for the amount of services that they actually consume. Many applications that process large workloads in parallel could potentially benefit from the elasticity offered by cloud computing providers. Unfortunately, current public cloud computing providers need to impose a strict limit on the amount of resources that a single user can simultaneously acquire. To address this limitation, we present an alternative approach to the construction of computational infrastructures to support cloud computing that is not based on traditional capacity planning. Inspired by Toyota’s Just in Time (JiT) philosophy, we introduce the concept of Just in Time Clouds to represent a new category of service in which the provider allocates resources only when actually demanded by customers and only while there is use for them. Exploring low scale outsourced resources, a JiT Cloud provider can increase its ability to offer IaaS in a more scalable way and with a virtually unlimited elasticity, since it is based on the discovery, federation and reselling of idle resources whose installation and operation costs are paid by a third party. We performed a proof of concept, on a network of Digital TV, to investigate the potential of utilization of outsourced resources with high granularity, high volatility and high dispersion for the construction of JiT Clouds with high throughput using a new architecture, called On-demand Distributed Computing Infrastructure (OddCI). Our results show that it is possible to build dynamic computing infrastructures based on computational resources placed in virtually the entire spectrum of low scale outsourced resources. In the most challenging scenarios, it was possible to obtain collective availability using isolated devices to deliver computational throughput with maximum losses of 10% under scenarios of up to 40% of volatility, caused by node unavailability. Considering the use of unconventional outsourced resources, as low cost Digital TV receivers , there was a significant difference in computational power compared with conventional low granularity devices, such as home PCs. However, this loss does not constitute an irreparable technical limitation, but only one circumstantial marketing aspect, that can be easily circumvented if a demand for more powerful devices is created.