Data (computer science)

• Random access memory (RAM) holds data that the CPU has direct access to. A CPU may only manipulate data within its processor registers or memory. This is as opposed to data storage, where the CPU must direct the transfer of data between the storage device (disk, tape...) and memory. RAM is an array of linear contiguous locations that a processor may read or write by providing an address for the read or write operation. The processor may operate on any location in memory at any time in any order. In RAM the smallest element of data is the binary bit. The capabilities and limitations of accessing RAM are processor specific. In general main memory is arranged as an array of locations beginning at address 0 (hexadecimal 0). Each location can store usually 8 or 32 bits depending on the computer architecture.

• Data keys need not be a direct hardware address in memory. Indirect, abstract and logical keys codes can be stored in association with values to form a data structure. Data structures have predetermined offsets (or links or paths) from the start of the structure, in which data values are stored. Therefore, the data key consists of the key to the structure plus the offset (or links or paths) into the structure. When such a structure is repeated, storing variations of the data values and the data keys within the same repeating structure, the result can be considered to resemble a table, in which each element of the repeating structure is considered to be a column and each repetition of the structure is considered as a row of the table. In such an organization of data, the data key is usually a value in one (or a composite of the values in several) of the columns.

• The tabular view of repeating data structures is only one of many possibilities. Repeating data structures can be organised hierarchically, such that nodes are linked to each other in a cascade of parent-child relationships. Values and potentially more complex data-structures are linked to the nodes. Thus the nodal hierarchy provides the key for addressing the data structures associated with the nodes. This representation can be thought of as an inverted tree. Modern computer operating system file systems are a common example; and XML is another.

• Data has some inherent features when it is sorted on a key. All the values for subsets of the key appear together. When passing sequentially through groups of the data with the same key, or a subset of the key changes, this is referred to in data processing circles as a break, or a control break. It particularly facilitates the aggregation of data values on subsets of a key.

• Until the advent of bulk non-volatile memory like flash, persistent data storage was traditionally achieved by writing the data to external block devices like magnetic tape and disk drives. These devices typically seek to a location on the magnetic media and then read or write blocks of data of a predetermined size. In this case, the seek location on the media, is the data key and the blocks are the data values. Early used raw disk data file-systems or disc operating systems reserved contiguous blocks on the disc drive for data files. In those systems, the files could be filled up, running out of data space before all the data had been written to them. Thus much unused data space was reserved unproductively to ensure adequate free space for each file. Later file-systems introduced partitions. They reserved blocks of disc data space for partitions and used the allocated blocks more economically, by dynamically assigning blocks of a partition to a file as needed. To achieve this, the file system had to keep track of which blocks were used or unused by data files in a catalog or file allocation table. Though this made better use of the disc data space, it resulted in fragmentation of files across the disc, and a concomitant performance overhead due additional seek time to read the data. Modern file systems reorganize fragmented files dynamically to optimize file access times. Further developments in file systems resulted in virtualization of disc drives i.e. where a logical drive can be defined as partitions from a number of physical drives.

• Retrieving a small subset of data from a much larger set may imply inefficiently searching through the data sequentially. Indexes are a way to copy out keys and location addresses from data structures in files, tables and data sets, then organize them using inverted tree structures to reduce the time taken to retrieve a subset of the original data. In order to do this, the key of the subset of data to be retrieved must be known before retrieval begins. The most popular indexes are the B-tree and the dynamic hash key indexing methods. Indexing is overhead for filing and retrieving data. There are other ways of organizing indexes, e.g. sorting the keys and using a binary search algorithm.

• Object-oriented programming uses two basic concepts for understanding data and software:

1. The taxonomic rank-structure of classes, which is an example of a hierarchical data structure; and
2. at run time, the creation of references to in-memory data-structures of objects that have been instantiated from a class library.

It is only after instantiation that an object of a specified class exists. After an object's reference is cleared, the object also ceases to exist. The memory locations where the object's data was stored are garbage and are reclassified as unused memory available for reuse.

• The advent of databases introduced a further layer of abstraction for persistent data storage. Databases use metadata, and a structured query language protocol between client and server systems, communicating over a computer network, using a two phase commit logging system to ensure transactional completeness, when saving data.

• Modern scalable and high-performance data persistence technologies, such as Apache Hadoop, rely on massively parallel distributed data processing across many commodity computers on a high bandwidth network. In such systems, the data is distributed across multiple computers and therefore any particular computer in the system must be represented in the key of the data, either directly, or indirectly. This enables the differentiation between two identical sets of data, each being processed on a different computer at the same time.