where:
DOIO	name of an I/O procedure
stream	number of the stream I/O will occur
mode	required operation
amount	amount of data to be transferred
destination	address into which data is to be input
(source	address from which data is to be output)
semaphore	address of semaphore request serviced which is to be
	signalled when I/O operation completes.

Functions & Characteristics of the I/O procedure.

• It must be re-entrant

• It should locate the Phy. Device identified by the stream (from stream descriptor list of calling process).

• It should validate the parameters

• Finally, it should initiate service of the request by assembling the parameters into an IORB and adding it to the device's device request queue (held in device descriptor)

• If the device is non-shareable, then the queue can contain IORBs from the one process only, otherwise, the queue can contain IORBs from any process.

• When the request has been added to the queue, the procedure signals a request pending.

• When I/O operation completes, device handler signals a request serviced.

procedure DOIO(stream, mode, amount, destination, semaphore);

begin lookup device in process descriptor;

check parameters against device characteristics;

if error then error exit;

assemble IORB;

add IORB to device request queue;

signal(request pending)

end;

• Each Device Class has its own handler.

• Device Handler waits for at least one request to be in the request queue.

• It picks the request at the top of the queue (unless entries are prioritised), extracts details of the request, initiates the appropriate I/O operation and waits on an `operation complete' semaphore.

• Just before entering ISR, device handler checks the status of the device.

• If request is for input, then the input data is translated before being transferred to destination.

• If request is for output, then data is translated before I/O procedure is called to output the data.

• If request is not for data transfer, then there is no data to translate or transfer.

• At the end of the operation, device handler signals `request serviced'.

• The semaphore `operation complete' is signalled by the ISR initiated by the Device Handler.

The interaction between the modules described above is summarised in the diagram below

Disadvantage of the operation described above

• Because the calling process is managing the synchronisation of semaphores, the programmer must include the mechanism in his/her code.

• This gives programmers the freedom to have the process continue execution in parallel with the service of the I/O request.

• However, there is room for error, because if the Wait is overlooked or put in the wrong place, then the process could try to use data that isn't there, or could continue after an error had occurred.

Restructuring the operation

• If the wait on `request serviced' is embedded in the I/O procedure, then
  • Programmers don't have to cater for the synchronisation
    
  • The process is automatically synchronised with the I/O operation
    
  • The synchronisation is `localised' (embedded in the OS)
    
  • The semaphore does not need to be a DOIO parameter
    

• So far operation described implies that every time a process request I/O operation, it is made unrunnable.

• Usually safe to assume that process will want to read/write more data while the stream is open.

• So it is efficient to read in more data than is requested on the assumption that the process will want to use other data that is logically close (`Locality').

• Data can be read into an Input Buffer - area of primary memory. Perform physical Input only when buffer is empty.

• Output - write data to output buffer, only writing it away to physical device when buffer is full.

• Processes then do not have to wait for I/O operation, because (most) reads/writes can be done in one operation.

Other Advantages

• Increases I/O & CPU usage, by increasing utilisation of I/O in parallel with CPU.

• Processes only made unrunnable if buffer is empty (on input) or full (on output).

• Number of times processes are made unrunnable can be reduced by adding more buffers (when buffer1 is full, write to buffer2 while buffer1 is emptied).

  Amending the I/O procedure.

• The number of buffers that a stream is to use must be specified when the stream is opened.

• IORB is only generated when buffer is full (empty), otherwise data is passed directly to (from) the process from (to) the buffer.

• When one IORB is generated, I/O procedure will generate enough others to fill (empty) the buffer.

• Procedure called by DOBUFFIO(stream, mode, destination).

• No need for Semaphore.

• Not all devices can be identified by name alone. This is true of sequential devices, where data can only be input or output sequentially, but it is not true of Random Access Devices.

• On a random access device it is possible to read data from or write data to logically separate areas in the same program. So a unique logical identification of each area must be given as well as the device name.

• The logical area used to store or retrieve data is called a file, and is identified by a file name.

• A device that supports files is called a file device, or file structured device.

• Buffering alone is not sufficient for non-shareable devices.

• E.g., all processes outputing to printer would be suspended while one process controls printer. Print job could last for more than 30 minutes. Unsatisfactory.

• Spooling (Simultaneous Peripheral Operations OnLine) is a more sophisticated form of buffering.

• Output (input) is stored on disk until device is free.

• Spooler moves data between disk andthe device.

• Device can then operate at optimum speed.

• Jobs in spool queue can be prioritised.

• Buffering can be used to smooth the transfer between disk and the device.

Advantages and Disadvantages

• Covered some I/O issues to derive concept of Virtual Device

• Looked at Device Handlers, to convert generic request to device specific request

• Looked at `localised' communication between system processes to achieve notion of apparent synchronisation between process execution and I/O operation

• Buffering

• Spooling

Next Lecture...