Chapter 9 Uniprocessor Scheduling - Wiiliem.ppt

Editor's Notes

• #1: “Operating Systems: Internal and Design Principles”, 8/e, by William Stallings, Chapter 9 “Uniprocessor Scheduling”.
• #2: In a multiprogramming system, multiple processes exist concurrently in main memory. Each process alternates between using a processor and waiting for some event to occur, such as the completion of an I/O operation. The processor or processors are kept busy by executing one process while the others wait. The key to multiprogramming is scheduling. In fact, four types of scheduling are typically involved (Table 9.1). One of these, I/O scheduling, is more conveniently addressed in Chapter 11, where I/O is discussed. The remaining three types of scheduling, which are types of processor scheduling, are addressed in this chapter and the next. This chapter begins with an examination of the three types of processor scheduling, showing how they are related. We see that long-term scheduling and mediumterm scheduling are driven primarily by performance concerns related to the degree of multiprogramming. These issues are dealt with to some extent in Chapter 3 and in more detail in Chapters 7 and 8. Thus, the remainder of this chapter concentrates on short-term scheduling and is limited to a consideration of scheduling on a uniprocessor system. Because the use of multiple processors adds additional complexity, it is best to focus on the uniprocessor case first, so that the differences among scheduling algorithms can be clearly seen.
• #3: The aim of processor scheduling is to assign processes to be executed by the processor or processors over time, in a way that meets system objectives, such as response time, throughput, and processor efficiency. In many systems, this scheduling activity is broken down into three separate functions: long-, medium-, and short-term scheduling. The names suggest the relative time scales with which these functions are performed.
• #4: Figure 9.1 relates the scheduling functions to the process state transition diagram (first shown in Figure 3.9b ). Long-term scheduling is performed when a new process is created. This is a decision whether to add a new process to the set of processes that are currently active. Medium-term scheduling is a part of the swapping function. This is a decision whether to add a process to those that are at least partially in main memory and therefore available for execution. Short-term scheduling is the actual decision of which ready process to execute next.
• #5: Figure 9.2 reorganizes the state transition diagram of Figure 3.9b to suggest the nesting of scheduling functions.
• #6: Scheduling affects the performance of the system because it determines which processes will wait and which will progress. This point of view is presented in Figure 9.3 , which shows the queues involved in the state transitions of a process. Fundamentally, scheduling is a matter of managing queues to minimize queuing delay and to optimize performance in a queuing environment.
• #7: The long-term scheduler determines which programs are admitted to the system for processing. Thus, it controls the degree of multiprogramming. Once admitted, a job or user program becomes a process and is added to the queue for the short-term scheduler. In some systems, a newly created process begins in a swapped-out condition, in which case it is added to a queue for the medium-term scheduler. In a batch system, or for the batch portion of a general-purpose operating system, newly submitted jobs are routed to disk and held in a batch queue. The long-term scheduler creates processes from the queue when it can. There are two decisions involved here. First, the scheduler must decide when the operating system can take on one or more additional processes. Second, the scheduler must decide which job or jobs to accept and turn into processes. The decision as to when to create a new process is generally driven by the desired degree of multiprogramming. The more processes that are created, the smaller is the percentage of time that each process can be executed (i.e., more processes are competing for the same amount of processor time). Thus, the long-term scheduler may limit the degree of multiprogramming to provide satisfactory service to the current set of processes. Each time a job terminates, the scheduler may decide to add one or more new jobs. Additionally, if the fraction of time that the processor is idle exceeds a certain threshold, the long-term scheduler may be invoked. The decision as to which job to admit next can be on a simple first-come-first- served (FCFS) basis, or it can be a tool to manage system performance. The criteria used may include priority, expected execution time, and I/O requirements. For example, if the information is available, the scheduler may attempt to keep a mix of processor-bound and I/O-bound processes. Also, the decision can depend on which I/O resources are to be requested, in an attempt to balance I/O usage. For interactive programs in a time-sharing system, a process creation request can be generated by the act of a user attempting to connect to the system. Timesharing users are not simply queued up and kept waiting until the system can accept them. Rather, the operating system will accept all authorized comers until the system is saturated, using some predefined measure of saturation. At that point, a connection request is met with a message indicating that the system is full and the user should try again later.
• #8: Medium-term scheduling is part of the swapping function. The issues involved are discussed in Chapters 3 , 7 , and 8 . Typically, the swapping-in decision is based on the need to manage the degree of multiprogramming. On a system that does not use virtual memory, memory management is also an issue. Thus, the swapping-in decision will consider the memory requirements of the swapped-out processes.
• #9: In terms of frequency of execution, the long-term scheduler executes relatively infrequently and makes the coarse-grained decision of whether or not to take on a new process and which one to take. The medium-term scheduler is executed somewhat more frequently to make a swapping decision. The short-term scheduler, also known as the dispatcher, executes most frequently and makes the fine-grained decision of which process to execute next. The short-term scheduler is invoked whenever an event occurs that may lead to the blocking of the current process or that may provide an opportunity to preempt a currently running process in favor of another. Examples of such events include • Clock interrupts • I/O interrupts • Operating system calls • Signals (e.g., semaphores)
• #10: The main objective of short-term scheduling is to allocate processor time in such a way as to optimize one or more aspects of system behavior. Generally, a set of criteria is established against which various scheduling policies may be evaluated. The commonly used criteria can be categorized along two dimensions. First, we can make a distinction between user-oriented and system-oriented criteria. User-oriented criteria relate to the behavior of the system as perceived by the individual user or process. An example is response time in an interactive system. Response time is the elapsed time between the submission of a request until the response begins to appear as output. This quantity is visible to the user and is naturally of interest to the user. We would like a scheduling policy that provides “good” service to various users. In the case of response time, a threshold may be defined, say two seconds. Then a goal of the scheduling mechanism should be to maximize the number of users who experience an average response time of two seconds or less. Other criteria are system oriented. That is, the focus is on effective and efficient utilization of the processor. An example is throughput, which is the rate at which processes are completed. This is certainly a worthwhile measure of system performance and one that we would like to maximize. However, it focuses on system performance rather than service provided to the user. Thus, throughput is of concern to a system administrator but not to the user population. Whereas user-oriented criteria are important on virtually all systems, system-oriented criteria are generally of minor importance on single-user systems. On a single-user system, it probably is not important to achieve high processor utilization or high throughput as long as the responsiveness of the system to user applications is acceptable.
• #11: Another dimension along which criteria can be classified is those that are performance related and those that are not directly performance related. Performance-related criteria are quantitative and generally can be readily measured. Examples include response time and throughput. Criteria that are not performance related are either qualitative in nature or do not lend themselves readily to measurement and analysis. An example of such a criterion is predictability. We would like for the service provided to users to exhibit the same characteristics over time, independent of other work being performed by the system. To some extent, this criterion can be measured by calculating variances as a function of workload. However, this is not nearly as straightforward as measuring throughput or response time as a function of workload.
• #12: Table 9.2 summarizes key scheduling criteria. These are interdependent, and it is impossible to optimize all of them simultaneously. For example, providing good response time may require a scheduling algorithm that switches between processes frequently. This increases the overhead of the system, reducing throughput. Thus, the design of a scheduling policy involves compromising among competing requirements; the relative weights given the various requirements will depend on the nature and intended use of the system. In most interactive operating systems, whether single user or time shared, adequate response time is the critical requirement. Because of the importance of this requirement, and because the definition of adequacy will vary from one application to another, the topic is explored further in Appendix G .
• #13: In many systems, each process is assigned a priority and the scheduler will always choose a process of higher priority over one of lower priority. Figure 9.4 illustrates the use of priorities. For clarity, the queuing diagram is simplified, ignoring the existence of multiple blocked queues and of suspended states (compare Figure 3.8a ). Instead of a single ready queue, we provide a set of queues, in descending order of priority: RQ0, RQ1, . . . , RQ n , with priority[RQ i ] > priority[RQ j ] for i > j . When a scheduling selection is to be made, the scheduler will start at the highest-priority ready queue (RQ0). If there are one or more processes in the queue, a process is selected using some scheduling policy. If RQ0 is empty, then RQ1 is examined, and so on. One problem with a pure priority scheduling scheme is that lower-priority processes may suffer starvation. This will happen if there is always a steady supply of higher-priority ready processes. If this behavior is not desirable, the priority of a process can change with its age or execution history. We will give one example of this subsequently.
• #14: Table 9.3 presents some summary information about the various scheduling policies that are examined in this subsection
• #15: The selection function determines which process, among ready processes, is selected next for execution. The function may be based on priority, resource requirements, or the execution characteristics of the process. In the latter case, three quantities are significant: w = time spent in system so far, waiting e = time spent in execution so far s = total service time required by the process, including e ; generally, this quantity must be estimated or supplied by the user For example, the selection function max[ w ] indicates an FCFS discipline.
• #16: The decision mode specifies the instants in time at which the selection function is exercised. There are two general categories: nonpreemptive and preemptive
• #17: Non-preemptive In this case, once a process is in the Running state, it continues to execute until it terminates or it blocks itself to wait for I/O or to request some operating system service. Preemptive: The currently running process may be interrupted and moved to the Ready state by the operating system. The decision to preempt may be performed when a new process arrives; when an interrupt occurs that places a blocked process in the Ready state; or periodically, based on a clock interrupt.
• #18: As we describe the various scheduling policies, we will use the set of processes in Table 9.4 as a running example. We can think of these as batch jobs, with the service time being the total execution time required. Alternatively, we can consider these to be ongoing processes that require alternate use of the processor and I/O in a repetitive fashion. In this latter case, the service times represent the processor time required in one cycle. In either case, in terms of a queuing model, this quantity corresponds to the service time.
• #19: For the example of Table 9.4 , Figure 9.5 shows the execution pattern for each policy for one cycle, and Table 9.5 summarizes some key results. First, the finish time of each process is determined. From this, we can determine the turnaround time. In terms of the queuing model, turnaround time (TAT) is the residence time T r , or total time that the item spends in the system (waiting time plus service time). A more useful figure is the normalized turnaround time, which is the ratio of turnaround time to service time. This value indicates the relative delay experienced by a process. Typically, the longer the process execution time, the greater is the absolute amount of delay that can be tolerated. The minimum possible value for this ratio is 1.0; increasing values correspond to a decreasing level of service.
• #20: The simplest scheduling policy is first-come-first-served (FCFS), also known as first-in-first-out (FIFO) or a strict queuing scheme. As each process becomes ready, it joins the ready queue. When the currently running process ceases to execute, the process that has been in the ready queue the longest is selected for running. FCFS performs much better for long processes than short ones. Another difficulty with FCFS is that it tends to favor processor-bound processes over I/O-bound processes. Consider that there is a collection of processes, one of which mostly uses the processor (processor bound) and a number of which favor I/O (I/O bound). When a processor-bound process is running, all of the I/O bound processes must wait. Some of these may be in I/O queues (blocked state) but may move back to the ready queue while the processor-bound process is executing. At this point, most or all of the I/O devices may be idle, even though there is potentially work for them to do. When the currently running process leaves the Running state, the ready I/O-bound processes quickly move through the Running state and become blocked on I/O events. If the processor-bound process is also blocked, the processor becomes idle. Thus, FCFS may result in inefficient use of both the processor and the I/O devices. FCFS is not an attractive alternative on its own for a uniprocessor system. However, it is often combined with a priority scheme to provide an effective scheduler. Thus, the scheduler may maintain a number of queues, one for each priority level, and dispatch within each queue on a first-come-first-served basis. We see one example of such a system later, in our discussion of feedback scheduling.
• #21: A straightforward way to reduce the penalty that short jobs suffer with FCFS is to use preemption based on a clock. The simplest such policy is round robin. A clock interrupt is generated at periodic intervals. When the interrupt occurs, the currently running process is placed in the ready queue, and the next ready job is selected on a FCFS basis. This technique is also known as time slicing , because each process is given a slice of time before being preempted. With round robin, the principal design issue is the length of the time quantum, or slice, to be used. If the quantum is very short, then short processes will move through the system relatively quickly. On the other hand, there is processing overhead involved in handling the clock interrupt and performing the scheduling and dispatching function. Thus, very short time quanta should be avoided. Round robin is particularly effective in a general-purpose time-sharing system or transaction processing system. One drawback to round robin is its relative treatment of processor-bound and I/O-bound processes. Generally, an I/O-bound process has a shorter processor burst (amount of time spent executing between I/O operations) than a processor-bound process. If there is a mix of processor-bound and I/O-bound processes, then the following will happen: An I/O-bound process uses a processor for a short period and then is blocked for I/O; it waits for the I/O operation to complete and then joins the ready queue. On the other hand, a processor-bound process generally uses a complete time quantum while executing and immediately returns to the ready queue. Thus, processor-bound processes tend to receive an unfair portion of processor time, which results in poor performance for I/O-bound processes, inefficient use of I/O devices, and an increase in the variance of response time.
• #22: One useful guide is that the time quantum should be slightly greater than the time required for a typical interaction or process function. If it is less, then most processes will require at least two time quanta. Figure 9.6 illustrates the effect this has on response time. Note that in the limiting case of a time quantum that is longer than the longest running process, round robin degenerates to FCFS.
• #23: Figure 9.5 and Table 9.5 show the results for our example using time quanta q of 1 and 4 time units. Note that process E, which is the shortest job, enjoys significant improvement for a time quantum of 1.
• #24: [HALD91] suggests a refinement to round robin that he refers to as a virtual round robin (VRR) and that avoids this unfairness. Figure 9.7 illustrates the scheme. New processes arrive and join the ready queue, which is managed on an FCFS basis. When a running process times out, it is returned to the ready queue. When a process is blocked for I/O, it joins an I/O queue. So far, this is as usual. The new feature is an FCFS auxiliary queue to which processes are moved after being released from an I/O block. When a dispatching decision is to be made, processes in the auxiliary queue get preference over those in the main ready queue. When a process is dispatched from the auxiliary queue, it runs no longer than a time equal to the basic time quantum minus the total time spent running since it was last selected from the main ready queue. Performance studies by the authors indicate that this approach is indeed superior to round robin in terms of fairness.
• #25: Another approach to reducing the bias in favor of long processes inherent in FCFS is the shortest process next (SPN) policy. This is a nonpreemptive policy in which the process with the shortest expected processing time is selected next. Thus, a short process will jump to the head of the queue past longer jobs. Figure 9.5 and Table 9.5 show the results for our example. Note that process E receives service much earlier than under FCFS. Overall performance is also significantly improved in terms of response time. However, the variability of response times is increased, especially for longer processes, and thus predictability is reduced. One difficulty with the SPN policy is the need to know or at least estimate the required processing time of each process. For batch jobs, the system may require the programmer to estimate the value and supply it to the operating system. If the programmer’s estimate is substantially under the actual running time, the system may abort the job. In a production environment, the same jobs run frequently, and statistics may be gathered. For interactive processes, the operating system may keep a running average of each “burst” for each process. A risk with SPN is the possibility of starvation for longer processes, as long as there is a steady supply of shorter processes. On the other hand, although SPN reduces the bias in favor of longer jobs, it still is not desirable for a time-sharing or transaction processing environment because of the lack of preemption. Looking back at our worst-case analysis described under FCFS, processes W, X, Y, and Z will still execute in the same order, heavily penalizing the short process Y.
• #26: The size of the coefficient as a function of its position in the expansion is shown in Figure 9.8 . The larger the value of , the greater is the weight given to the more recent observations. For α= 0.8, virtually all of the weight is given to the four most recent observations, whereas for α = 0.2, the averaging is effectively spread out over the eight or so most recent observations. The advantage of using a value of close to 1 is that the average will quickly reflect a rapid change in the observed quantity. The disadvantage is that if there is a brief surge in the value of the observed quantity and it then settles back to some average value, the use of a large value of α will result in jerky changes in the average.
• #27: Figure 9.9 compares simple averaging with exponential averaging (for two different values of α). Here (Figure 9.9a), the observed value begins at 1, grows gradually to a value of 10, and then stays there.
• #28: While here (Figure 9.9b), the observed value begins at 20, declines gradually to 10, and then stays there. In both cases, we start out with an estimate of S1 = 0. This gives greater priority to new processes. Note that exponential averaging tracks changes in process behavior faster than does simple averaging and that the larger value of α results in a more rapid reaction to the change in the observed value.
• #29: The shortest remaining time (SRT) policy is a preemptive version of SPN. In this case, the scheduler always chooses the process that has the shortest expected remaining processing time. When a new process joins the ready queue, it may in fact have a shorter remaining time than the currently running process. Accordingly, the scheduler may preempt the current process when a new process becomes ready. As with SPN, the scheduler must have an estimate of processing time to perform the selection function, and there is a risk of starvation of longer processes. SRT does not have the bias in favor of long processes found in FCFS. Unlike round robin, no additional interrupts are generated, reducing overhead. On the other hand, elapsed service times must be recorded, contributing to overhead. SRT should also give superior turnaround time performance to SPN, because a short job is given immediate preference to a running longer job. Note that in our example ( Table 9.5 ), the three shortest processes all receive immediate service, yielding a normalized turnaround time for each of 1.0.
• #30: In Table 9.5 , we have used the normalized turnaround time, which is the ratio of turnaround time to actual service time, as a figure of merit. For each individual process, we would like to minimize this ratio, and we would like to minimize the average value over all processes. In general, we cannot know ahead of time what the service time is going to be, but we can approximate it, either based on past history or some input from the user or a configuration manager. Thus, our scheduling rule becomes the following: when the current process completes or is blocked, choose the ready process with the greatest value of R . This approach is attractive because it accounts for the age of the process. While shorter jobs are favored (a smaller denominator yields a larger ratio), aging without service increases the ratio so that a longer process will eventually get past competing shorter jobs. As with SRT and SPN, the expected service time must be estimated to use highest response ratio next (HRRN).
• #31: Figure 9.10 illustrates the feedback scheduling mechanism by showing the path that a process will follow through the various queues. This approach is known as multilevel feedback , meaning that the operating system allocates the processor to a process and, when the process blocks or is preempted, feeds it back into one of several priority queues.
• #32: In this section, we make use of basic queuing formulas, with the common assumptions of Poisson arrivals and exponential service times. First, we make the observation that any such scheduling discipline that chooses the next item to be served independent of service time obeys the following relationship: (see formula) In particular, a priority-based scheduler, in which the priority of each process is assigned independent of expected service time, provides the same average turnaround time and average normalized turnaround time as a simple FCFS discipline. Furthermore, the presence or absence of preemption makes no differences in these averages. With the exception of round robin and FCFS, the various scheduling disciplines considered so far do make selections on the basis of expected service time. Unfortunately, it turns out to be quite difficult to develop closed analytic models of these disciplines. However, we can get an idea of the relative performance of such scheduling algorithms, compared to FCFS, by considering priority scheduling in which priority is based on service time.
• #33: If scheduling is done on the basis of priority and if processes are assigned to a priority class on the basis of service time, then differences do emerge. Table 9.6 shows the formulas that result when we assume two priority classes, with different service times for each class. In the table, refers to the arrival rate. These results can be generalized to any number of priority classes. Note that the formulas differ for nonpreemptive versus preemptive scheduling. In the latter case, it is assumed that a lower-priority process is immediately interrupted when a higher-priority process becomes ready.
• #34: As an example, let us consider the case of two priority classes, with an equal number of process arrivals in each class and with the average service time for the lower-priority class being five times that of the upper priority class. Thus, we wish to give preference to shorter processes. Figure 9.11 shows the overall result. By giving preference to shorter jobs, the average normalized turnaround time is improved at higher levels of utilization. As might be expected, the improvement is greatest with the use of preemption. Notice, however, that overall performance is not much affected.
• #35: However, significant differences emerge when we consider the two priority classes separately. Figure 9.12 shows the results for the higher-priority, shorter processes. For comparison, the upper line on the graph assumes that priorities are not used but that we are simply looking at the relative performance of that half of all processes that have the shorter processing time. The other two lines assume that these processes are assigned a higher priority. When the system is run using priority scheduling without preemption, the improvements are significant. They are even more significant when preemption is used.
• #36: Figure 9.13 shows the same analysis for the lower-priority, longer processes. As expected, such processes suffer a performance degradation under priority scheduling.
• #37: Some of the difficulties of analytic modeling are overcome by using discrete-event simulation, which allows a wide range of policies to be modeled. The disadvantage of simulation is that the results for a given “run” only apply to that particular collection of processes under that particular set of assumptions. Nevertheless, useful insights can be gained. The results of one such study are reported in [FINK88]. The simulation involved 50,000 processes with an arrival rate of λ = 0.8 and an average service time of T s = 1. Thus, the assumption is that the processor utilization is ρ = λT s = 0.8. Note, therefore, that we are only measuring one utilization point. To present the results, processes are grouped into service-time percentiles, each of which has 500 processes. Thus, the 500 processes with the shortest service time are in the first percentile; with these eliminated, the 500 remaining processes with the shortest service time are in the second percentile; and so on. This allows us to view the effect of various policies on processes as a function of the length of the process. Figure 9.14 shows the normalized turnaround time.
• #38: Figure 9.15 shows the average waiting time. Looking at the turnaround time, we can see that the performance of FCFS is very unfavorable, with one-third of the processes having a normalized turnaround time greater than 10 times the service time; furthermore, these are the shortest processes. On the other hand, the absolute waiting time is uniform, as is to be expected because scheduling is independent of service time. The figures show round robin using a quantum of one time unit. Except for the shortest processes, which execute in less than one quantum, round robin yields a normalized turnaround time of about five for all processes, treating all fairly. Shortest process next performs better than round robin, except for the shortest processes. Shortest remaining time, the preemptive version of SPN, performs better than SPN except for the longest 7% of all processes. We have seen that, among nonpreemptive policies, FCFS favors long processes and SPN favors short ones. Highest response ratio next is intended to be a compromise between these two effects, and this is indeed confirmed in the figures. Finally, the figure shows feedback scheduling with fixed, uniform quanta in each priority queue. As expected, FB performs quite well for short processes.
• #39: All of the scheduling algorithms discussed so far treat the collection of ready processes as a single pool of processes from which to select the next running process. This pool may be broken down by priority but is otherwise homogeneous. However, in a multiuser system, if individual user applications or jobs may be organized as multiple processes (or threads), then there is a structure to the collection of processes that is not recognized by a traditional scheduler. From the user’s point of view, the concern is not how a particular process performs but rather how his or her set of processes, which constitute a single application, performs. Thus, it would be attractive to make scheduling decisions on the basis of these process sets. This approach is generally known as fair-share scheduling. Further, the concept can be extended to groups of users, even if each user is represented by a single process. For example, in a time-sharing system, we might wish to consider all of the users from a given department to be members of the same group. Scheduling decisions could then be made that attempt to give each group similar service. Thus, if a large number of people from one department log onto the system, we would like to see response time degradation primarily affect members of that department rather than users from other departments. The term fair share indicates the philosophy behind such a scheduler. Each user is assigned a weighting of some sort that defines that user’s share of system resources as a fraction of the total usage of those resources. In particular, each user is assigned a share of the processor. Such a scheme should operate in a more or less linear fashion, so that if user A has twice the weighting of user B, then in the long run, user A should be able to do twice as much work as user B. The objective of a fair-share scheduler is to monitor usage to give fewer resources to users who have had more than their fair share and more to those who have had less than their fair share.
• #40: Figure 9.16 is an example in which process A is in one group and processes B and C are in a second group, with each group having a weighting of 0.5. Assume that all processes are processor bound and are usually ready to run. All processes have a base priority of 60. Processor utilization is measured as follows: The processor is interrupted 60 times per second; during each interrupt, the processor usage field of the currently running process is incremented, as is the corresponding group processor field. Once per second, priorities are recalculated. In the figure, process A is scheduled first. At the end of one second, it is preempted. Processes B and C now have the higher priority, and process B is scheduled. At the end of the second time unit, process A has the highest priority. Note that the pattern repeats: the kernel schedules the processes in order: A, B, A, C, A, B, and so on. Thus, 50% of the processor is allocated to process A, which constitutes one group, and 50% to processes B and C, which constitute another group.
• #41: In this section we examine traditional UNIX scheduling, which is used in both SVR3 and 4.3 BSD UNIX. These systems are primarily targeted at the time-sharing interactive environment. The scheduling algorithm is designed to provide good response time for interactive users while ensuring that low-priority background jobs do not starve. Although this algorithm has been replaced in modern UNIX systems, it is worthwhile to examine the approach because it is representative of practical time-sharing scheduling algorithms. The scheduling scheme for SVR4 includes an accommodation for real-time requirements, and so its discussion is deferred to Chapter 10 . The traditional UNIX scheduler employs multilevel feedback using round robin within each of the priority queues. The system makes use of one-second preemption. That is, if a running process does not block or complete within one second, it is preempted. Priority is based on process type and execution history.
• #42: The priority of each process is recomputed once per second, at which time a new scheduling decision is made. The purpose of the base priority is to divide all processes into fixed bands of priority levels. The CPU and nice components are restricted to prevent a process from migrating out of its assigned band (assigned by the base priority level). These bands are used to optimize access to block devices (e.g., disk) and to allow the operating system to respond quickly to system calls.
• #43: An example of process scheduling is shown in Figure 9.17 . Processes A, B, and C are created at the same time with base priorities of 60 (we will ignore the nice value). The clock interrupts the system 60 times per second and increments a counter for the running process. The example assumes that none of the processes block themselves and that no other processes are ready to run. Compare this with Figure 9.16 .
• #44: Summary of Chapter 9.