A batch system is one in which jobs are bundled together with the instructions necessary to allow them to be processed without intervention.
Often jobs of a similar nature can be bundled together to further increase economy
The basic physical layout of the memory of a batch job computer is shown below:
-------------------------------------- | | | Monitor (permanently resident) | | | -------------------------------------- | | | User Space | | (compilers, programs, data, etc.) | | | --------------------------------------
The monitor is system software that is responsible for interpreting and carrying out the instructions in the batch jobs. When the monitor started a job, it handed over control of the entire computer to the job, which then controlled the computer until it finished.
A sample of several batch jobs might look like:
$JOB user_spec ; identify the user for accounting purposes $FORTRAN ; load the FORTRAN compiler source program cards $LOAD ; load the compiled program $RUN ; run the program data cards $EOJ ; end of job $JOB user_spec ; identify a new user $LOAD application $RUN data $EOJ
Often magnetic tapes and drums were used to store intermediate data and compiled programs.
As mentioned above, one of the major shortcomings of early batch systems was that there was no protection scheme to prevent one job from adversely affecting other jobs.
The solution to this was a simple protection scheme, where certain memory (e.g. where the monitor resides) were made off-limits to user programs. This prevented user programs from corrupting the monitor.
To keep user programs from reading too many (or not enough) cards, the hardware was changed to allow the computer to operate in one of two modes: one for the monitor and one for the user programs. IO could only be performed in monitor mode, so that IO requests from the user programs were passed to the monitor. In this way, the monitor could keep a job from reading past it's on $EOJ card.
To prevent an infinite loop, a timer was added to the system and the $JOB card was modified so that a maximum execution time for the job was passed to the monitor. The computer would interrupt the job and return control to the monitor when this time was exceeded.
One difficulty with simple batch systems is that the computer still needs to read the the deck of cards before it can begin to execute the job. This means that the CPU is idle (or nearly so) during these relatively slow operations.
Since it is faster to read from a magnetic tape than from a deck of cards, it became common for computer centers to have one or more less powerful computers in addition to there main computer. The smaller computers were used to read a decks of cards onto a tape, so that the tape would contain many batch jobs. This tape was then loaded on the main computer and the jobs on the tape were executed. The output from the jobs would be written to another tape which would then be removed and loaded on a less powerful computer to produce any hardcopy or other desired output.
It was a logical extension of the timer idea described above to have a timer that would only let jobs execute for a short time before interrupting them so that the monitor could start an IO operation. Since the IO operation could proceed while the CPU was crunching on a user program, little degradation in performance was noticed.
Since the computer can now perform IO in parallel with computation, it became possible to have the computer read a deck of cards to a tape, drum or disk and to write out to a tape printer while it was computing. This process is called SPOOLing: Simultaneous Peripheral Operation OnLine.
Spooling batch systems were the first and are the simplest of the multiprogramming systems.
One advantage of spooling batch systems was that the output from jobs was available as soon as the job completed, rather than only after all jobs in the current cycle were finished.
As machines with more and more memory became available, it was possible to extend the idea of multiprogramming (or multiprocessing) as used in spooling batch systems to create systems that would load several jobs into memory at once and cycle through them in some order, working on each one for a specified period of time.
-------------------------------------- | Monitor | | (more like a operating system) | -------------------------------------- | User program 1 | -------------------------------------- | User program 2 | -------------------------------------- | User program 3 | -------------------------------------- | User program 4 | --------------------------------------
At this point the monitor is growing to the point where it begins to resemble a modern operating system. It is responsible for:
As a simple, yet common example, consider a machine that can run two jobs at once. Further, suppose that one job is IO intensive and that the other is CPU intensive. One way for the monitor to allocate CPU time between these jobs would be to divide time equally between them. However, the CPU would be idle much of the time the IO bound process was executing.
A good solution in this case is to allow the CPU bound process (the background job) to execute until the IO bound process (the foreground job) needs some CPU time, at which point the monitor permits it to run. Presumably it will soon need to do some IO and the monitor can return the CPU to the background job.
Back in the days of the "bare" computers without any operating system to speak of, the programmer had complete access to the machine. As hardware and software was developed to create monitors, simple and spooling batch systems and finally multiprogrammed systems, the separation between the user and the computer became more and more pronounced.
Users, and programmers in particular, longed to be able to "get to the machine" without having to go through the batch process. In the 1970s and especially in the 1980s this became possible two different ways.
The first involved timesharing or timeslicing. The idea of multiprogramming was extended to allow for multiple terminals to be connected to the computer, with each in-use terminal being associated with one or more jobs on the computer. The operating system is responsible for switching between the jobs, now often called processes, in such a way that favored user interaction. If the context-switches occurred quickly enough, the user had the impression that he or she had direct access to the computer.
Interactive processes are given a higher priority so that when IO is requested (e.g. a key is pressed), the associated process is quickly given control of the CPU so that it can process it. This is usually done through the use of an interrupt that causes the computer to realize that an IO event has occurred.
It should be mentioned that there are several different types of time sharing systems. One type is represented by computers like our VAX/VMS computers and UNIX workstations. In these computers entire processes are in memory (albeit virtual memory) and the computer switches between executing code in each of them. In other types of systems, such as airline reservation systems, a single application may actually do much of the timesharing between terminals. This way there does not need to be a different running program associated with each terminal.
The second way that programmers and users got back at the machine was the advent of personal computers around 1980. Finally computers became small enough and inexpensive enough that an individual could own one, and hence have complete access to it.
A real-time computer is one that execute programs that are guaranteed to have an upper bound on tasks that they carry out. Usually it is desired that the upper bound be very small. Examples included guided missile systems and medical monitoring equipment. The operating system on real-time computers is severely constrained by the timing requirements.
Dedicated computers are special purpose computers that are used to perform only one or more tasks. Often these are real-time computers and include applications such as the guided missile mentioned above and the computer in modern cars that controls the fuel injection system.
A multiprocessor computer is one with more than one CPU. The category of multiprocessor computers can be divided into the following sub-categories:
shared memory multiprocessors have multiple CPUs, all with access to the same memory. Communication between the the processors is easy to implement, but care must be taken so that memory accesses are synchronized.
distributed memory multiprocessors also have multiple CPUs, but each CPU has it's own associated memory. Here, memory access synchronization is not a problem, but communication between the processors is often slow and complicated.
networked systems consist of multiple computers that are networked together, usually with a common operating system and shared resources. Users, however, are aware of the different computers that make up the system.
distributed systems also consist of multiple computers but differ from networked systems in that the multiple computers are transparent to the user. Often there are redundant resources and a sharing of the workload among the different computers, but this is all transparent to the user.
$Id: history.html,v 1.3 2000/01/07 17:03:38 senning Exp $
These notes are based on a set of notes by Prof. R. Bjork, Gordon College and the textbooks Operating System Concepts by Silberschatz and Galvin, Addison-Wesley, 1998 and Operating Systems: Design and Implementation by Tanenbaum and Woodhull, Prentice-Hall, 1997.