Buffering is a technique used to match a small-but-steady process (e.g. a program that reads or writes one line at a time) to a large-block process (e.g. disk I/O).

Disk I/O has two problematic features:

• A whole disk block (e.g. 4096 bytes) must be read or written at a time.

• Disk access is slow (e.g. 8 milliseconds).

An I/O buffer is an array, the same size as a disk block, that is used to collect data. The application program removes data from the block (or adds data to it) until the block is empty (full), at which time a new block is read from disk (written to disk).

If there are R Reduce tasks, each Map task will have R output buffers, one for each Reduce task. When an output buffer becomes full, it is written to disk. When the Map task is finished, it sends the file names of its R files to the Master.

Contents    Page-10    Prev    Next    Page+10    Index