I want to preface this by saying, I have a feeling that this idea will not work the way I'm imagining, but I'm not sure why. Its likely I'm making some sort of false assumption about the way the internet works.
Lets say server A has a file of size 1024 kb. This file is split up into 1024 packets to be sent over a network. Server A sends these 1024 packets to server B via TCP. As soon as B receives a packet it sends it back to A and vice versa. If any 3rd party C makes a request to A for the data, it would make a copy of each packet it receives from B and send to to C and B.
In this scheme, server A could delete its file once it has sent all the packets to B, freeing up disk space. Servers A and B only need to store a single packet at any given time and the rest of the packets would be "juggled" in the network.
Is it really possible to store data by "juggling" it through a network? Am I underestimating the overhead in receiving and sending packets?
EDIT: This is assuming 100% network reliability, which is probably a completely unrealistic assumption.
A network does have some sort of "storage" capacity, and it is usually measured by the "Bandwidth-delay product". (Think network as a pipe, then its storage-capacity, the amount of data(water) that the pipe can hold is the volume of that pipe, the area-length product.) If you use unreliable protocols such as UDP to do the "cyclic sending", the extra data that exceeds the capacity will simply lose. If you use TCP to do it, after filling up the "pipe", the sending will then fill up the internal sending buffer on the OS, and then block.