Скачать или смотреть Cloud Computing Fail: Why Scientists Hate the On-Demand Model

Cloud computing's business model is failing scientists, leading to higher costs and unreliable access to resources.
Key problems include short, infrequent scientific simulations clashing with persistent service models from vendors like AWS or Google Cloud. Researchers need specialized hardware for bioinformatics or astrophysics but face risks with preemptible instances that can crash entire MPI-based simulations.
Unlike businesses, scientific projects are often finite and grant-funded, lacking the long-term purchasing power to get discounts. Institutions don't strategize for better deals, and credits from vendors don't guarantee future partnerships.
Even on-demand elasticity fails when compute isn't available, charging scientists for idle time while waiting for resources that never arrive. The paper calls for better collaboration between cloud providers and scientists to create cost models that support discovery without sacrificing profitability.
Explore how modern science relies on cloud for modeling, yet struggles with budgets and availability. Discuss alternatives like on-prem supercomputers and the need for guaranteed resource allocation.