Fast AmigaDOS I/O


by Martin Taillefer


Reading and writing data is crucial to most applications and is, in
many cases, a major bottleneck.  Using the Amiga's sophisticated file
system architecture can help reduce, and sometimes eliminate, the
time spent waiting for I/O to complete.  This article presents six
small routines that can greatly improve an application's I/O
performance.

Typically, an application processes a file in the following manner:

Step 1: Open the file.
Step 2: Read some data (with the DOS library's Read() function).
Step 3: Process that data.
Step 4: Repeat steps 2 and 3 until the application is finished
        processing the file.
Step 5: Close file.

This sequence of steps is effective, but it does have a potential
bottleneck.  Whenever the application reads some data using the DOS
Read() function, the Amiga has to put that task to sleep and ask the
file system to fetch the data.  The file system then starts up the
disk hardware and reads the data.  After the file system finishes
reading the data, the operating system wakes up the application.

The problem is step 2.  While the file system is busy reading data
from the disk, the application is idle, waiting for the DOS I/O in
Read() to complete.  A more sophisticated application would initiate
an asynchronous read, allowing the application to continue to do some
other important chore while the file system is busy reading.  If all
goes well, the file system will be finished with the asynchronous
read by the time the application is finished with its chore, so the
application will not have to wait for any DOS I/O to complete before
the application can access data.

Using the routines presented in this article, an application
processes a file in the following manner:

Step 1:  Open the file with OpenAsync().  This function opens the
         file and, if the file is opened for reading, OpenAsync()
         asks the file system to start reading data, asynchronously.
Step 2:  Read some data with ReadAsync().  If the asynchronous read
         request that OpenAsync() sent has not completed, ReadAsync()
         will put the application to sleep until that request
         returns. Ideally, the read will have returned, so the
         application won't have to wait. ReadAsync() will also
         initiate a new asynchronous read so new file data is ready
         when the application needs it.
Step 3:  Process the file data.
Step 4:  Repeat steps 2 and 3 until the application processes
         all its file data.
Step 5:  Close the file with CloseAsync().

Immediately after opening the file, OpenAsync() sends a request to
the file system to get it reading data in the background.  If all
goes well, by the time the application gets around to reading the
first byte of data, the file system has already copied the data into
memory.  That means the application doesn't need to wait and can
immediately start processing the data.  As soon as the application
starts processing data from the file using ReadAsync(), ReadAsync()
sends out a second request to the file system to fill up a second
buffer.  Once the application is done processing the first buffer, it
starts processing the second one.  When this happens, the file system
starts filling up the first buffer again with new data.  This process
continues until the application has read all of its data.  This
technique is known as ``double-buffered asynchronous I/O''.

The set of functions presented below offer high-performance I/O using
the technique described above.  The interface is very similar to
standard AmigaDOS files.  These routines enable full asynchronous
read/write of any file.

These functions are especially useful on an Amiga with a DMA (Direct
Memory Access) hard drive.  DMA makes it possible to transfer data to
memory at the same time the CPU is busy executing a task's
instructions.  A DMA data transfer is truly parallel, so, under
normal conditions, the CPU is operating at full speed, unaffected by
the DMA transfer.  This parallelism is what makes the set of
accompanying routines so efficient.  They exploit the fact that the
Amiga can transfer an application's data while the application is
busy processing other data.

Although these asynchronous routines make disk I/O much faster, they
do have an important limitation.  The routines do not support seeking
into a file.

 Async.doc 
 ASyncExample.c 
 ASyncIO.c 
 ASyncIO.h