Me running a @dgraphlabs ingestor (not using their initial import tools) streaming from my SSD to my slow HDD with 100 go routines:
127 entries per second
SSD -> (same) SSD:
2229 entries per second
So yea lolhttps://twitter.com/generativist/status/1196879332389445632 …
This ingestor routine isn't loading synthetic data, it's loading the real albeit pre-serialized stream as fast as it can (given the fixed number of goroutines). So the data access patterns should at least afford a decent estimate.