This isn’t well-specified enough to be a real challenge. They say it can be a server - but not too beefy of a server. What does that even mean? If I put eight NVMe SSDs in a 128-core server, is that too beefy? What about 64, 16 cores?
Can I know (or bound) the number of orders or products in advance to preallocate? Can I design the dataset myself with certain assumptions (e.g. sorted with respect to time)? Can I bound certain aspects of the dataset (e.g. orders must not contain more than 255 products, orders always contain the prices of everything, etc.)?
Latency isn’t apparently a factor - so if I’m processing 1B records, do we care how quickly it gets done? If not, I’ll just stream the data off to a GPU and get the results later?
Can I know (or bound) the number of orders or products in advance to preallocate? Can I design the dataset myself with certain assumptions (e.g. sorted with respect to time)? Can I bound certain aspects of the dataset (e.g. orders must not contain more than 255 products, orders always contain the prices of everything, etc.)?
Latency isn’t apparently a factor - so if I’m processing 1B records, do we care how quickly it gets done? If not, I’ll just stream the data off to a GPU and get the results later?