Algorithms and Data Structures in External Memory
Vojin G. Oklobdzija · 2001
In large applications, data sets are often too massive to fit completely inside the computer’s internal memory. The resulting input/output (or I/O) communication between fast internal memory and slower external memory (such as disks) can be a major performance bottleneck. For example, loading a register takes on the order of a nanosecond (10 s), and accessing internal memory takes tens of nanoseconds, but the latency of accessing data from a disk is several milliseconds (10 s), which is about one million times slower.