Amazon, Microsoft, and Google cloud services make preemptable 'spot market' compute cycles available at significant discounts (up to 90%), but the availability and pricing of the spot instances varies based on market demand. A similar situation occurs on a smaller scale in the Duke Compute Cluster where scavenger/common pool instance availability varies based on demand, and low priority jobs can be terminated to make way for high priority work.
A team of students will work with the Office of Information Technology and other Duke stakeholders to develop analytic tools to help decide if it is worthwhile to run on a spot market instance, which would help inform (and potentially automate) cloud-busting large scale compute jobs and arbitrage the cloud compute market. The analytic tools would look at historic and up-to-the minute market trends to model likely market behavior.
Student Project Managers/Facilitators