Issue link: https://read.uberflip.com/i/1187300
Amazon Web Services – HPC Lens AWS Well-Architected Framework Page 40 For these applications, the run efficiency of a case typically falls off at higher core counts. A cost versus turnaround-time curve can be created if many cases of similar type and size will be run. Curves are specific to both the case type and application as scaling depends significantly on the ratio of computational to network requirements. Larger workloads are capable of scaling further than smaller workloads. With an understanding of the cost versus turnaround-time tradeoff, time- sensitive workloads can be run more quickly on more cores, while cost savings can be achieved by running on fewer cores and at maximum efficiency. Workloads can also fall somewhere in between when you want to balance time sensitivity and cost sensitivity. Definition There are four best practice areas for cost optimization in the cloud: • Cost-effective resources • Matching supply and demand • Expenditure awareness • Optimizing over time The matching supply and demand, expenditure awareness, and optimizing over time categories are vital and well described in the AWS Well-Architected Framework whitepaper. They will not be described in this paper because the practices in the AWS Well-Architected Framework paper do not require modification for HPC workloads. Best Practices Cost-Effective Resources Using the appropriate instances and resources for your system is key to cost management. The instance choice may increase or decrease the overall cost of running an HPC workload. For example, a tightly coupled HPC workload might take five hours to run on a cluster of several smaller servers, while a cluster of fewer and larger servers may cost double per hour but compute the result in one hour, saving money overall. The choice of storage can also impact cost. Therefore, it is important to consider the potential tradeoff between job turnaround and cost optimization and to consider testing workloads with different instance sizes and storage options to optimize cost.