local data access

Terms from Artificial Intelligence: humans at the heart of algorithms

The glossary is being gradually proof checked, but currently has many typos and misspellings.

Local data access is when the data required for code to execute is physically close to it, in the sense of on the same processor, or at least a close network. This is an issue for big data processing especially on cloud computing platforms, where data may be stored in multiple computers often in physically remote data centres. The goal is often to perform as much processing as possible locally without moving data around. Where the code is accessing multiple data items, it is important that these are then stored physically close to one another. This may be because there are intrinsic commonality about data so they can be permanently stored close to one another, or they may be brought together for the purposes of processing. An example all the former is data about a single person, and maybe summary information about their friends, in a social network which can be stored as a single record or data item. An example of the latter is MapReduce, which uses a common hash value to bring items that need to be processed together by the reduce step on the same processor.

Used in Chap. 8: page 111