To cloud or not to cloud ?
The Bumbershoot Corporation's biology research center produces 600 GB of new data for every wet lab experiment. Assume the data generated can be easily parallelized, with a negligible overhead.
(a) Suppose Amazon Web Services (AWS) sells CPU hours at the price of $0.10 per hour per Elastic Compute Cloud (EC2) instance, where each instance takes 2 hours to process 1 GB of the experimental data. The data-transfer fee is $0.15/GB. What is the price Bumbershoot will need to pay for processing an experiment using the EC2 service?
(b) Suppose the data-transfer rate from the research center to AWS is 20 Mbps. What is the total time required to transmit and process the experimental data using the EC2 service?
(c) The Bumbershoot Corporation has 24 computers itself, each taking 2 hours to process 1 GB of data. Suppose the overall cost (including electricity, software, hardware, etc) is $15 per computer per experiment. What is the total amount of time and cost required to process the experiment? Will Bumbershoot Corporation be willing to use the EC2 service?
(d) We saw in (b) and (c) that the data transmission time can be a problem in cloud computing. Can you think of a way to overcome this obstacle, so that Bumbershoot can still process the experiment using the EC2 service within a day?