To take full advantage of IO-Lite, application programs can use an extended I/O application programming interface (API) that is based on buffer aggregates. This section briefly describes this API. A complete discussion of the API can be found in the technical report .
IOL_read and IOL_write form the core of the interface (see Figure 2). These operations supersede the standard UNIX read and write operations. (The latter operations are maintained for backward compatibility.) Like their predecessors, the new operations can act on any UNIX file descriptor. All other file descriptor related UNIX systems calls remain unchanged.
The new IOL_read operation returns a buffer aggregate (IOL_Agg) containing at most the amount of data specified as an argument. Unlike the POSIX read, IOL_read may always return less data than requested. The IOL_write operation replaces the data in an external data object with the contents of the buffer aggregate passed as an argument.
The effects of IOL_read and IOL_write operations are atomic with respect to other IOL_write operations concurrently invoked on the same descriptor. That is, an IOL_read operations yields data that either reflects all or none of the changes affected by a concurrent IOL_write operation on the same file descriptor. The data returned by a IOL_read is effectively a ``snapshot'' of the data contained in the object associated with the file descriptor.
Additional IO-Lite system calls allow the creation and deletion of IO-Lite allocation pools. A version of IOL_read allows applications to specify an allocation pool, such that the system places the requested data into IO-Lite buffers from that pool. Applications that manage multiple I/O streams with different access control lists use this operation. The (IOL_Agg) buffer aggregate abstract data type supports a number of operations for creation, destruction, duplication, concatenation and truncation as well as data access.
Implementations of language-specific runtime I/O libraries, like the ANSI C stdio library, can be converted to use the new API internally. Doing so reduces data copying without changing the library's API. As a result, applications that perform I/O using these standard libraries can enjoy some performance benefits merely by re-linking them with the new library.