Short name: Aerial Insights
Long name: Performance evaluation of cost aware aggressive scheduling approaches for multi datacenter computing
Company: Aerial Insights S.L. | aerial-insights.co
Call: F4Fp-01-S (see call details)
Proposal number: F4Fp-01-02-S
SUMMARY REMARKS & TESTBEDS
Aerial Insights is an online service that provides image processing, storage and analytics to drone pilots. Once a flight mission is completed, pilot scan upload the raw images as they are produced by the drone and our automated workflow pre-process them, assembles maps, analyses their contents and hosts the results so they are available via an online multiplatform interface. Our service cost a fraction of other alternatives which means pilots, usually freelances and SME, can save 15K-20K€/year. Aggressive pricing is key to the success of the service, which is mainly dependent on the price of licenses for 3rd party packages and cloud computing resources. More details of the platform can be found at www.aerial-insights.co.
Our current production deployment builds upon the infrastructure of a single IaaS provider. While technically convenient, this reduces opportunities to take advantage of pricing from alternative providers with lower usage tiers, reduced fees or even bid for spare computing resources. To compute a single map, a high-end computer may be fully assigned for several hours. Since customers tend to be flexible with delivery times, there are plenty of opportunities to use innovative scheduling, data centers in different geographical locations as well as discounts and special tiers during off-peak hours.The target of this experiment is testing a series of approaches for job scheduling, process monitoring, IaaS resource identification as well as aggressive rescheduling across data centers. The algorithms and supporting technology are implemented for the most part.
They have been tested under controlled conditions but these scenarios are too simplistic as they cannot easily simulate real world factors such as load generated by hundreds of users around the world, data and job transfers between data centers, heterogeneous computing devices or the performance of networks/Internet. Fed4fire is a great opportunity to take advantage of a high end facility to put such systems under stress and determine how they will react under multiple real world situations.