In the world of wire and wireless things, information is being shoveled around in astounding amounts. This also includes devices which interface "beneath" the web, using IP but not HTTP, so that transactions between machines can be optimal for that pair or set of machines involved, and can be composed of data like video streams, audio streams, telemetry, meta-programs and so on.
But the state of all things on all wired and wireless signal networks is from such an enormous quantity of data that the usefulness of it is very low. You cannot recognize everything at once, all the time. A human mind is good at making a story from the random puffs of data that enter via the eyes and ears and all other senses. The story cannot be the simultaneous stories of all possible paths of analysis. Some smaller set is always used for the useful stories about the world of information.
Big Data is a phrase that means a lot of different things; it depends on the point of view used to refer to the Big Data. Storing it is a whole complicated pack of dogs. Correlating it is the job of a pack of genies who can create other genies to help them. Reaction to it requires accumulation of experiences regarding what happens when certain patterns in the combinations of correlations occur. Delegating each sub-combination to a another reactor might go for very deep levels because of the huge number of states each sub-combination may itself consist of.
In concrete terms, use Laptops with 8 processors as a unit. An entire laptop is sometimes loaded with enormous amounts of data, although it has only 8 processors. Each of the 8 processors may also subdivide certain data into sub-processes that are either emulated by the 8 processor system as a whole in a multitasking manner, or are hardware like graphics cells that can process hundreds of data nodes in parallel, perhaps 128.
In another construction of machinery, such as a forest of server nodes, the actual number of unique processors can be in the millions. This can take some problems and solve them in seconds instead of years or centuries. Such machines are not without cost, since there is a tremendous complexity in assuring all states in all the processors is synchronized as well as parallel executed. If there are trillions of nodes, it may take weeks to load all the states of each node just to get ready to start solving the original problem.
So, if the number of humans on the planet is in the billions, and there are billions of cameras in cell phones, millions of CCTV nodes and web cameras. Each has a huge amount of data, perhaps mostly useless. There are huge numbers of comparisons made amongst the mountains of data by the billions of computer operations it takes. There are trillions of unique comparisons necessary to process just the nodes, and each node has gigabytes of data. Assuming intelligent compression of all that data is made (not so far) then it might take far fewer nodes to process it all later.
I imagine there may actually be agencies that collect and sift this Big Data, using a myriad of tools to help harvest and gain whatever advantage it gives them to catch crooks or keep us less than free, whichever is closer to infinity.