RIOT: a Novel Stochastic Method for Rapidly Configuring Cloud-Based Workflows

08/27/2017 ∙ by Jianfeng Chen, et al. ∙ 0

Traditional tools for configuring cloud services can run much slower than the workflows they are trying to optimize. For example, in the case studies reported here, we find cases where (using traditional methods) it takes hours to find ways to make a workflow terminate in tens of seconds. Such slow optimizers are a poor choice of tools for reacting to changing operational environmental conditions. Hence, they are unsuited for cloud services that support rapidly changing workflows, e.g., scientific workflows or workflows from the media or telecommunication industries. To solve this problem, this paper presents RIOT (Randomized Instance Order Types), a new configuration tool. RIOT has a very low optimization overhead-- often, less than 10% of the system runtime, especially for every complex workflow. Instead of simulating many configurations, RIOT uses a novel surrogate sampling method to quickly find promising solutions. As shown by this paper, RIOT achieves comparable results to the other approaches but does so in a fraction of the time.



There are no comments yet.


page 1

page 2

page 3

page 4

This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.