As an example, to read a simple mailing address from a modest 10GB (10 billion bytes) file storing customer information on a file storage service like IPFS, the entire 10GB file would need to be downloaded first to your local machine. Then, the file would have to be searched for the right information.
Imagine a 1 Gbps (1 Gbps = 1 billion gigabits per second) download rate, which is a reasonable fast and modern network connection. The file’s 10 billion bytes is 80 billion bits (there are 8 bits in a byte), so downloading (ie: reading) this file alone would take 80 seconds every single time that “database” is accessed.
Now imagine a real database (ie: Bluzelle) that has the same 10GB of data in it. That same 32 byte (256 bits) mailing address will now take less than 100 milliseconds to be read from the network, because the Bluzelle database knows exactly what is being asked for and ONLY responds with the needed data.
https://blog.bluzelle.com/how-is-bluzelle-different-from-ipfs-filecoin-storj-sia-and-ethereums-swarm-1d1d792658f8