Kernel implementation

If you want higher performance for your applications, such as database applications, use the Kernel Implementation. The Kernel Implementation contains the following caveats:

• It supports Asynchronous I/O only on raw slices of hard disks.

• It supports the same interface but must be used with two additional interfaces, aio_memlock and the AIO_RAW flag.

  • The aio_memlock function must be called to lock the I/O buffer in memory before any other Asynchronous I/O operations can be performed.

  • The AIO_RAW flag has to be set in the aiocb control block so that the Asynchronous I/O operations will be performed by the kernel.

• It can be used on systems that only support raw disk slices such as SVR4.2.

• It supports lio_listio, but does not support aio_suspend and aio_cancel.

Accessing raw disk slices

The aio_flags member of the aiocb control block must be set to AIO_RAW if you want to access raw disk slices. The following example shows you how to set this flag:

   aiocb.aio_flags = AIO_RAW;

Using the asynchronous I/O memory lock

The Asynchronous I/O interface uses aio_memlock to perform memory lock operations. See aio_memlock(3aio). Before starting an Asynchronous I/O request, the application must lock an area of memory to be used for the I/O buffer. The size of the locked area is determined by the number of requests that may be outstanding at any point in time. Only the I/O buffer area must be locked. Once this buffer has been locked with aio_memlock, it can not be resized or unlocked. When the application exits, the memory is unlocked automatically.