Process Scheduling 國立中正大學 資訊工程研究所 羅習五 老師
Outline OS schedulers Unix scheduling Linux scheduling Linux 2.4 scheduler Linux 2.6 scheduler O(1) scheduler CFS
Introduction preemptive & cooperative multitasking A multitasking operating system is one that can simultaneously interleave execution of more than one process. Multitasking operating systems come in two flavors: cooperative multitasking and preemptive multitasking. Linux provides preemptive multitasking MAC OS 9 and earlier being the most notable cooperative multitasking .
UNIX Scheduling Policy Scheduling policy determines what runs when fast process response time (low latency) maximal system utilization (high throughput) Processes classification: I/O-bound processes: spends much of its time submitting and waiting on I/O requests Processor-bound processes: spend much of their time executing code Unix variants tends to favor I/O-bound processes, thus providing good process response time
Linux scheduler – Process Priority Linux’s priority-based scheduling Rank processes based on their worth and need for processor time. processes with a higher priority also receive a longer timeslice. Both the user and the system may set a process's priority to influence the scheduling behavior of the system. Dynamic priority-based scheduling Begins with an initial base priority Then enables the scheduler to increase or decrease the priority dynamically to fulfill scheduling objectives. E.g., a process that is spending more time waiting on I/O will receive an elevated dynamic priority.
Linux scheduler – Priority Ranges Two separate priority ranges. nice value, from -20 to +19 with a default of 0. Larger nice values correspond to a lower priority. (you are being nice to the other processes on the system). real-time priority, by default range from 0 to 99. All real-time processes are at a higher priority than normal processes. Linux implements real-time priorities in accordance with POSIX standards on the matter.
scheduler – priority
Timeslice The timeslice is the numeric value that represents how long a task can run until it is pre-empted. too short => large overhead of switching process too long => poor interactive response Linux’s CFS scheduler does not directly assign timeslices to processes. CFS assigns processes a proportion of the processor. the amount of processor time that a process receives is a function of the load of the system
2.4 scheduler
2.4 scheduler - SMP busy run queue busy
2.4 scheduler - SMP IDLE search & estimate run queue busy
2.4 scheduler - SMP busy run queue busy
2.4 scheduler Non-preemptible kernel Round-robin Set p->need_resched if schedule() should be invoked at the ‘next opportunity‘ (kernel => user mode). Round-robin task_struct->counter: number of clock ticks left to run in this scheduling slice, decremented by a timer.
2.4 scheduler Check if schedule() was invoked from interrupt handler (due to a bug) and panic if so. Use spin_lock_irq() to lock ‘runqueue_lock’ Check if a task is ‘runnable’ in TASK_RUNNING state in TASK_INTERRUPTIBLE state and a signal is pending Examine the ‘goodness’ of each process Context switch
2.4 scheduler – ‘goodness’ ‘goodness’: identifying the best candidate among all processes in the runqueue list. ‘goodness’ = 0: the entity has exhausted its quantum. 0 < ‘goodness’ < 1000: the entity is a conventional process/thread that has not exhausted its quantum; a higher value denotes a higher level of goodness.
2.4 scheduler – ‘goodness’ (to improve multithreading performance) if (p->mm == prev->mm) return p->counter + p->priority + 1; else return p->counter + p->priority; A small bonus is given to the task p if it shares the address space with the previous task.
2.4 scheduler - SMP Examine the processor field of the processes and gives a consistent bonus (that is PROC_CHANGE_PENALTY, usually 15) to the process that was last executed on the ‘this_cpu’ CPU.
Recalculating Timeslices (kernel 2.4) Problems: Can take a long time. Worse, it scales O(n) for n tasks on the system. Recalculation must occur under some sort of lock protecting the task list and the individual process descriptors. This results in high lock contention. Nondeterminism is a problem with deterministic real-time programs.
Processes classification Definition: I/O-bound processes: spends much of its time submitting and waiting on I/O requests Processor-bound processes: spend much of their time executing code Linux tends to favor I/O-bound processes, thus providing good process response time How to classify processes?
Time quantum = 0 (CPU bound) tq=0 tq=0 tq=0 High priority tasks tq=0 Time quantum ≠ 0 I/O bound tq≠0 tq≠0 tq≠0 tq≠0 tq=???
Scheduling policy & time_quantumnew = bonusI/O + timestatic time_quantumnew = time_quantumold/2 + time_quantum_table[static_priority] & dynamic_priority ≈ time_quantumnew
Time quantum = 0 (CPU bound) tq=0 tq=0 tq=0 tq=? Time quantum ≠ 0 I/O bound tq≠0 tq≠0 tq≠0 tq≠0 tq=???
Time quantum = 0 (CPU bound) tq=0 tq=0 tq=0 tq=? Time quantum ≠ 0 I/O bound tq≠0 tq≠0 tq≠0 tq≠0 tq=???
tq=0 tq=? tq≠0 Time quantum ≠ 0 I/O bound tq≠0 tq≠0 tq≠0 tq=???
2.4 scheduler - performance The algorithm does not scale well It is inefficient to re-compute all dynamic priorities at once. The predefined quantum is too large for high system loads (for example: a server) I/O-bound process boosting strategy is not optimal a good strategy to ensure a short response time for interactive programs, but… some batch programs with almost no user interaction are I/O-bound.
2.6 scheduler
task migration (put + pull) 2.6 scheduler run queue task migration (put + pull) run queue
2.6 scheduler – User Preemption User preemption can occur When returning to user-space from a system call When returning to user-space from an interrupt handler
2.6 scheduler – Kernel Preemption The Linux kernel is a fully preemptive kernel. It is possible to preempt a task at any point, so long as the kernel is in a state in which it is safe to reschedule. “safe to reschedule”: kernel does not hold a lock The Linux design: adding of a preemption counter, preempt_count, to each process's thread_info This count increments once for each lock that is acquired and decrements once for each lock that is released Kernel preemption can also occur explicitly, when a task in the kernel blocks or explicitly calls schedule(). no additional logic is required to ensure that the kernel is in a state that is safe to preempt!
Kernel Preemption Kernel preemption can occur When an interrupt handler exits, before returning to kernel-space When kernel code becomes preemptible again If a task in the kernel explicitly calls schedule() If a task in the kernel blocks (which results in a call to schedule())
O(1) & CFS scheduler 2.5 ~ 2.6.22: O(1) scheduler Time complexity: O(1) Using “run queue” (an active Q and an expired Q) to realize the ready queue 2.6.23~present: Completely Fair Scheduler (CFS) Time complexity: O(log n) the ready queue is implemented as a red-black tree
2.6 scheduler – O(1)
O(1) scheduler Implement fully O(1) scheduling. Every algorithm in the new scheduler completes in constant-time, regardless of the number of running processes. (Since the 2.5 kernel). Implement perfect SMP scalability. Each processor has its own locking and individual runqueue. Implement improved SMP affinity. Attempt to group tasks to a specific CPU and continue to run them there. Only migrate tasks from one CPU to another to resolve imbalances in runqueue sizes. Provide good interactive performance. Even during considerable system load, the system should react and schedule interactive tasks immediately. Provide fairness. No process should find itself starved of timeslice for any reasonable amount of time. Likewise, no process should receive an unfairly high amount of timeslice. Optimize for the common case of only one or two runnable processes, yet scale well to multiple processors, each with many processes.
The Priority Arrays Each runqueue contains two priority arrays (defined in kernel/sched.c as struct prio_array) Active array: all tasks with timeslice left. Expired array: all tasks that have exhausted their timeslice. Priority arrays provide O(1) scheduling. Each priority array contains one queue of runnable processors per priority level. The priority arrays also contain a priority bitmap used to efficiently discover the highest-priority runnable task in the system.
The Linux O(1) scheduler algorithm
The Priority Arrays Each runqueuecontains two priority arrays (defined in kernel/sched.cas struct prio_array) Active array: all tasks with timesliceleft. Expired array: all tasks that have exhausted their timeslice. Priority arrays provide O(1) scheduling. Each priority array contains one queue of runnable processors per priority level. The priority arrays also contain a priority bitmap used to efficiently discover the highest-priority runnable task in the system.
Each runqueue contains two priority arrays – active and expired. Each of these priority arrays contains a list of tasks indexed according to priority runqueue Priority queue (0-139) active expired
Time quantum ≈ 1/priority Linux assigns higher-priority tasks longer time-slice runqueue Time quantum ≈ 1/priority tsk1 tsk2 tsk3 expired active
Linux chooses the task with the highest priority from the active array for execution. runqueue tsk1 tsk2 tsk3 expired active
runqueue tsk1 Round-robin tsk2 tsk3 expired active
runqueue tsk1 Round-robin tsk3 tsk2 expired active
runqueue tsk1 tsk2 tsk3 expired active
Most tasks have dynamic priorities that are based on their “nice” value (static priority) plus or minus 5 Interactivity of a task ≈ 1/sleep_time runqueue dynPrio = staticPrio + bonus bonus = -5 ~ +5 bonus ≈ 1/sleep_time tsk1 tsk3 tsk2 tsk3 I/O bound expired active
When all tasks have exhausted their time slices, the two priority arrays are exchanged! runqueue tsk1 tsk3 tsk2 expired active
The O(1) scheduling algorithm sched_find_first_bit() 1 1 1 tsk1 tsk3 tsk2
The O(1) scheduling algorithm Insert O(1) Remove O(1) 1 1 1 find first set bit O(1)
find first set bit O(1) static inline unsigned long __ffs (unsigned long word) { int num = 0; #if BITS_PER_LONG == 64 if ((word & 0xffffffff) == 0) { num += 32; word >>= 32; } #endif if ((word & 0xffff) == 0) { num += 16; word >>= 16; if ((word & 0xff) == 0) { num += 8; word >>= 8; if ((word & 0xf) == 0) { num += 4; word >>= 4; if ((word & 0x3) == 0) { num += 2; word >>= 2; if ((word & 0x1) == 0) num += 1; return num;
2.6 scheduler - CFS
2.6 scheduler – CFS The inventor of the CFS set himself a goal of devising a scheduler capable of the fair division of available CPU power among all tasks. If one had an ideal multitasking computer capable of concurrent execution on N processes then every process would get exactly 1/N-th of its available CPU power.
2.6 scheduler – CFS Classical schedulers compute time slices for each process in the system and allow them to run until their time slice/quantum is used up. After that, all process need to be recalculated. CFS considers only the wait time of a process The task with the most need for CPU time is scheduled.
2.6 scheduler - CFS
2.6 scheduler – CFS The inventor of the CFS set himself a goal of devising a scheduler capable of the fair devision of available CPU power among all tasks. If one had an ideal multitasking computer capable of concurrent execution on N processes then every process would get exactly 1/N-th of its available CPU power.
2.6 scheduler – CFS Classical schedulers compute time slices for each process in the system and allow them to run until their time slice/quantum is used up. After that, all process need to be recalculated. CFS considers only the wait time of a process The task with the most need for CPU time is scheduled.
2.6 scheduler – CFS
vruntime The vruntime variable stores the virtual runtime of a process, which is the actual runtime normalized by the number of runnable processes. The virtual runtime’s units are nanoseconds and therefore vruntime is decoupled from the timer tick. The virtual runtime is used to help us approximate the “ideal multitasking processor” that CFS is modeling.
updating vruntime update_curr() calculates the execution time of the current process and stores that value in delta_exec. It then passes that runtime to __update_curr(), which weights the time by the number of runnable processes.
Process Selection When CFS is deciding what process to run next, it picks the process with the smallest vruntime. CFS uses a red-black tree to manage the list of runnable processes and efficiently find the process with the smallest vruntime.
Adding Processes to the Tree This would occur when a process becomes runnable (wakes up) or is first created via fork(). se->vruntime += cfs_rq->min_vruntime; update_curr(cfs_rq); account_entity_enqueue(cfs_rq, se); update_stats_enqueue(cfs_rq, se); __enqueue_entity(cfs_rq, se);
2.6 scheduler – issues Different priority levels for tasks (i.e., nice values) must be taken into account Tasks must not be switched too often because a context switch has a certain overhead.
2.6 scheduler – issues time slice is a CPU timeslice that the task deserves, period is the epoch length task_load – the weighted task loading cfs rq load – the weight of the fair queue.
Scheduling policies for I/O-bound tasks Every tasks (i.e. i/o-bound tasks) that is waken up gets the virtual runtime equal to the smallest virtual runtime among the tasks in the queue. OR smallest_virtual - epsilon Run the i/o tasks first
2.6 scheduler – fields in the task_struct
2.6 scheduler – fields in the task_struct prio and normal_prio indicate the dynamic priorities, static_prio the static priority of a process. The static priority is the priority assigned to the process when it was started. The normal_prio & prio denote a priority that is computed based on the static priority and the scheduling policy of the process.
2.6 scheduler – fields in the task_struct cpus_allowed is a bit field used on multiprocessor systems to restrict the CPUs on which a process may run. setaffinity() getaffinity()
2.6 scheduler – priority
prio_to_weight[i] = prio_to_weight[i] ×1.25 2.6 scheduler – priority kernel/sched.c static const int prio_to_weight[40] = { /* -20 */ 88761, 71755, 56483, 46273, 36291, /* -15 */ 29154, 23254, 18705, 14949, 11916, /* -10 */ 9548, 7620, 6100, 4904, 3906, /* -5 */ 3121, 2501, 1991, 1586, 1277, /* 0 */ 1024, 820, 655, 526, 423, /* 5 */ 335, 272, 215, 172, 137, /* 10 */ 110, 87, 70, 56, 45, /* 15 */ 36, 29, 23, 18, 15, }; prio_to_weight[i] = prio_to_weight[i] ×1.25
2.6 scheduler – priority
Summary The concept of OS schedulers Maximize throughput. This is what system administrators care about. How to maximize throughput (CPU & I/O). What is the major drawback of Linux 2.4 scheduler To understand the pros and cons of Linux 2.6 schedulers O(1) CFS
期末報告 題目 task scheduling main memory management virtual memory & virtual memory space virtual file system & btrfs interrupt service routine & device drivers virtual machine (software approaches or hardware approaches)