I’m building a new cloud product that quickly processes large amounts of scientific data. Our largest customer dataset so far is about 3,000 tables, each 40 to 80 MB and totaling 150 GB, which we aim to process in 10 seconds or less. Each table can be processed independently, so we parallelize heavily to reach this goal — our deployment uses 1000 vCPUs or more as needed. The tricky part is rapidly