Vectored I/O

In computing, vectored I/O, also known as scatter/gather I/O, is a method of input and output by which a single procedure-call sequentially writes data from multiple buffers to a single data stream or reads data from a data stream to multiple buffers. The buffers are given in a vector of buffers. Scatter/gather refers to the process of gathering data from, or scattering data into, the given set of buffers. Vectored I/O can operate synchronously or asynchronously. The main reasons for using vectored I/O are efficiency and convenience.

There are several usages for vectored I/O:

  • Atomicity: If the particular vectored I/O implementation supports atomicity, a process can write into or read from a set of buffers to or from a file without risk that another thread or process might perform I/O on the same file between the first process' reads or writes, thereby corrupting the file or compromising the integrity of the input
  • Concatenating output: An application that wants to write non-sequentially placed data in memory can do so in one vectored I/O operation. For example, writing a fixed-size header and its associated payload data that are placed non-sequentially in memory can be done by a single vectored I/O operation without first concatenating the header and the payload to another buffer
  • Efficiency: One vectored I/O read or write can replace many ordinary reads or writes, and thus save on the overhead involved in syscalls
  • Splitting input: When reading data that is in a format that defines a fixed-size header, one can use a vector of buffers in which the first buffer is the size of that header; and the second buffer will contain the data associated with the header

Standards bodies document the applicable functions readv and writev in POSIX 1003.1-2001 and the Single UNIX Specification version 2. The Windows API has analogous functions ReadFileScatter and WriteFileGather; however unlike the POSIX functions they require the alignment of each buffer on a memory page. Windows Sockets provide separate WSASend and WSARecv functions without this requirement.

While working directly with a vector of buffers can be significantly harder than working with a single buffer, there are often higher-level APIs for working efficiently that eliminate the problem.