1. Load Balance in Linux 2.6.32 Load balancing Sung-joon Choi Real-Time Operating Systems Lab. Seoul National University 2011-09-15

2. 2 Contents  Load balancing Purpose Definition General cases Active load balancing Passive load balancing Special cases Execution of a new task CPU’s shut down or intentionally being IDLE Limitation Load Balance in Linux 2.6.32

3. 3 Load Balancing  Purpose 시스템에 코어 수보다 많은 수의 작업 (task) 이 있는 한, 모든 코어가 IDLE 상태 없이 수행하도록 조절  Mechanism 코어 간에 작업량 차이가 크지 않도록 조절  Definition Load balancing SMP 구조에서 각 코어가 균등한 작업량 (load) 을 가지도록 조절 하는 것 Load 코어의 run-queue 가 갖는 모든 task 들의 weight 를 더한 값 Load Balance in Linux 2.6.32

4. 4 Load Balancing  Definition (cont.) Idlest run-queue A run-queue that has the minimum load among the cores Busiest run-queue A run-queue that has the maximum value which is scale factor “load / (core’s power)” 모든 코어의 power 가 동일하다면 maximum load 를 갖는 코어 의 run-queue 를 의미한다 이종의 프로세서를 사용하는 시스템이라면 각 코어의 power 가 다를수도 있다. 일반적으로 power 는 capacity 또는 작업수행능력을 의미한다. Load Balance in Linux 2.6.32

5. 5 Contents  Load balancing Purpose Definition General cases (mainly focused part) Active load balancing Passive load balancing Special cases Execution of a new task CPU’s shut down or intentionally being IDLE Limitation Load Balance in Linux 2.6.32

6. 6 General Cases  Active Load Balancing Load Balance in Linux 2.6.32 Run-queueRun-queueRun-queueRun-queue Current task Core 0 Core 1 Task 1 Task 2 Task 4 Task 3 Task 5 Run-queueRun-queueRun-queueRun-queue Current task Core 0 Core 1 Task 1 Task 2 Task 4 Task 3 Run-queueRun-queueRun-queueRun-queue Current task Core 0 Core 1 Task 1 Task 2 Task 4 Task 3 Run-queueRun-queueRun-queueRun-queue Current task Core 0 Core 1 Task 1 Task 2 Task 3 READYREADYRUNNINGRUNNING Going to DEAD Core 1 is going to IDLE Run-queue is empty Task migration Task 2 (Assumption: all tasks have same weight) (Assumption: all tasks have same weight)

7. 7 Active Load Balancing  Implementation When a task is going to end up its execution time  do_exit() Sets task’s state to “TASK_DEAD”  schedule() –In back-end procedure, if a core’s state is IDLE, it calls “idle_balance()” –  idle_balance() »To pull a task on the busiest core’s run-queue, it calls “load_balance()” »  load_balance() »Does a task migration General Cases

8. 8 Active Load Balancing  Drawback Active load balancing 으로도 충분히 load balancing 을 달성 할 수 있지만 코어 간 작업량 차이가 큰 상황인데도 각 태스크의 수행시간 이 길어서 IDLE 상태를 갖게 되는 코어가 한동안 없다면, 단 기간 내 load balancing 의 목적을 달성할 수 없다. 이 상황을 피하기 위해서 주기적인 조절이 필요하다 General Cases

9. 9  Passive(Periodic) load balancing Load Balance in Linux 2.6.32 Run-queueRun-queueRun-queueRun-queue Current task Core 0 Core 1 Task 1 Task 2 Task 5 Task 4 Task 6 Run-queueRun-queueRun-queueRun-queue Current task Core 0 Core 1 Task 1 Task 2 Task 5 Task 3 Run-queueRun-queueRun-queueRun-queue Current task Core 0 Core 1 Task 1 Task 2 Task 5 Task 3 Run-queueRun-queueRun-queueRun-queue Current task Core 0 Core 1 Task 4 Task 6 Task 3 Busiest run-queue Idlest run-queue Task 2 Task 3 For a long time, there is no IDLE core Task 4 Task 6 Periodic check If there is big gap of load between cores, it is uncomfortable Task 4 Task 6 Task 1 Task 2 Task 5 Task migration READYREADYRUNNINGRUNNING Going to DEAD (Assumption: all tasks have same weight) (Assumption: all tasks have same weight)

10. 10 Passive Load Balancing  Triggered by scheduler_tick() Tick value is compared with a parameter “next_balance” which is the time to do load balancing Each run-queue has “next_balance” If a core takes the active load balancing, the parameter is set to 1 second after If a core takes the passive load balancing, the parameter is set to 1 minute after 1 초와 1 분의 차이는 IDLE 상태를 밸런싱했던 코어는 다시 IDLE 상태가 되기 쉽기 때문에 곧바로 밸런싱을 해주기 위한 것  Executed by bottom-half handler A softirq named “SCHED_SOFTIRQ” is handled by “run_rebalance_domains()” General Cases

11. 11 Passive Load Balancing  Implementation – start load balance General Cases Timer interrupt invokes “scheduler_tick()” If the tick value is equal to or greater than parameter “next_balance”, Busiest run-queue Idlest run-queue READYREADYRUNNINGRUNNING (Assumption: all tasks have same weight) (Assumption: all tasks have same weight) Run-queueRun-queueRun-queueRun-queue Current task Core 0 Core 1 Task 1 Task 2 Task 5 Task 7 Task 3 Task 6 Next_balanceNext_balanceNext_balanceNext_balance Task 4

12. 12 Passive Load Balancing  Implementation – step1 General Cases If the tick value is equal to or greater than parameter “next_balance”, Step1: raises a softirq “SCHED_SOFTIRQ” to kernel If the tick value is equal to or greater than parameter “next_balance”, Step1: raises a softirq “SCHED_SOFTIRQ” to kernel Softirq table …… ?????? SCHED_SOFTIRQSCHED_SOFTIRQ Busiest run-queue Idlest run-queue READYREADYRUNNINGRUNNING (Assumption: all tasks have same weight) (Assumption: all tasks have same weight) Run-queueRun-queueRun-queueRun-queue Current task Core 0 Core 1 Task 1 Task 2 Task 5 Task 7 Task 3 Task 6 Next_balanceNext_balanceNext_balanceNext_balance Task 4

13. 13 Passive Load Balancing  Implementation – step2 General Cases Run-queueRun-queueRun-queueRun-queue Current task Core 0 Core 1 Task 1 Task 2 ksoftirqdksoftirqd Task 5 Task 7 Busiest run-queue Idlest run-queue Task 3 READYREADYRUNNINGRUNNING (Assumption: all tasks have same weight) (Assumption: all tasks have same weight) If the tick value is equal to or greater than parameter “next_balance”, Step1: raises a softirq “SCHED_SOFTIRQ” to kernel Step2: finds the idlest run-queue to invoke a kernel thread “ksoftirqd” If the tick value is equal to or greater than parameter “next_balance”, Step1: raises a softirq “SCHED_SOFTIRQ” to kernel Step2: finds the idlest run-queue to invoke a kernel thread “ksoftirqd” Softirq table …… ?????? SCHED_SOFTIRQSCHED_SOFTIRQ Task 6 Next_balanceNext_balanceNext_balanceNext_balance Task 4

14. 14 Passive Load Balancing  Implementation – step3 General Cases Run-queueRun-queueRun-queueRun-queue Current task Core 0 Core 1 Task 1 Task 2 ksoftirqdksoftirqd Task 5 Task 7 Busiest run-queue Idlest run-queue Task 3 READYREADYRUNNINGRUNNING (Assumption: all tasks have same weight) (Assumption: all tasks have same weight) If the tick value is equal to or greater than parameter “next_balance”, Step1: raises a softirq “SCHED_SOFTIRQ” to kernel Step2: finds the idlest run-queue to invoke a kernel thread “ksoftirqd” If the tick value is equal to or greater than parameter “next_balance”, Step1: raises a softirq “SCHED_SOFTIRQ” to kernel Step2: finds the idlest run-queue to invoke a kernel thread “ksoftirqd” Softirq table …… ?????? SCHED_SOFTIRQSCHED_SOFTIRQ Task 6 Next_balanceNext_balanceNext_balanceNext_balance Task 4 Step3: the thread executes a function “do_ksoftirqd()” that picks a softirq and calls its handler function picks a softirq and calls its handler function Step3: the thread executes a function “do_ksoftirqd()” that picks a softirq and calls its handler function picks a softirq and calls its handler function run_rebalance_domains()run_rebalance_domains() Handler function (bottom-half handler)

15. 15 Passive Load Balancing  Implementation – step4 General Cases Run-queueRun-queueRun-queueRun-queue Current task Core 0 Core 1 Task 1 Task 2 ksoftirqdksoftirqd Task 5 Task 7 Busiest run-queue Idlest run-queue Task 3 READYREADYRUNNINGRUNNING (Assumption: all tasks have same weight) (Assumption: all tasks have same weight) If the tick value is equal to or greater than parameter “next_balance”, Step1: raises a softirq “SCHED_SOFTIRQ” to kernel Step2: finds the idlest run-queue to invoke a kernel thread “ksoftirqd” If the tick value is equal to or greater than parameter “next_balance”, Step1: raises a softirq “SCHED_SOFTIRQ” to kernel Step2: finds the idlest run-queue to invoke a kernel thread “ksoftirqd” Softirq table …… ?????? SCHED_SOFTIRQSCHED_SOFTIRQ Task 6 Next_balanceNext_balanceNext_balanceNext_balance Task 4 Step3: the thread executes a function “do_ksoftirqd()” that picks a softirq and calls its handler function picks a softirq and calls its handler function Step4: the handler function finds the busiest run-queue to pull a task Step3: the thread executes a function “do_ksoftirqd()” that picks a softirq and calls its handler function picks a softirq and calls its handler function Step4: the handler function finds the busiest run-queue to pull a task run_rebalance_domains()run_rebalance_domains() Handler function (bottom-half handler)

16. 16 Passive Load Balancing  Implementation – step5 General Cases Run-queueRun-queueRun-queueRun-queue Current task Core 0 Core 1 Task 1 Task 2 ksoftirqdksoftirqd Task 5 Task 7 Busiest run-queue Idlest run-queue Task 3 READYREADYRUNNINGRUNNING (Assumption: all tasks have same weight) (Assumption: all tasks have same weight) If the tick value is equal to or greater than parameter “next_balance”, Step1: raises a softirq “SCHED_SOFTIRQ” to kernel Step2: finds the idlest run-queue to invoke a kernel thread “ksoftirqd” If the tick value is equal to or greater than parameter “next_balance”, Step1: raises a softirq “SCHED_SOFTIRQ” to kernel Step2: finds the idlest run-queue to invoke a kernel thread “ksoftirqd” Softirq table …… ?????? SCHED_SOFTIRQSCHED_SOFTIRQ Task 6 Next_balanceNext_balanceNext_balanceNext_balance Task 4 Step3: the thread executes a function “do_ksoftirqd()” that picks a softirq and calls its handler function picks a softirq and calls its handler function Step4: the handler function finds the busiest run-queue to pull a task Step5: task migration Step3: the thread executes a function “do_ksoftirqd()” that picks a softirq and calls its handler function picks a softirq and calls its handler function Step4: the handler function finds the busiest run-queue to pull a task Step5: task migration run_rebalance_domains()run_rebalance_domains() Handler function (bottom-half handler) Task 4

17. 17 Passive Load Balancing  Implementation General Cases Run-queueRun-queueRun-queueRun-queue Current task Core 0 Core 1 Task 1 Task 2 Task 5 Task 6 Task 3 Task 4 Next_balanceNext_balanceNext_balanceNext_balance Task 7 Run-queueRun-queueRun-queueRun-queue Current task Core 0 Core 1 Task 1 Task 2 Task 5 Task 7 Task 3 Task 6 Next_balanceNext_balanceNext_balanceNext_balance Task 4

18. 18 Passive Load Balancing  Drawback This algorithm has large overhead The algorithm should check the maximum and minimum load out of all cores And, if a current core is not the idlest one, –The kernel thread “ksoftirqd” should be enqueued to the idlest run- queue of other core and waken up –Also, a current task of the target core that has the idlest run-queue is preempted by “ksoftirqd” Tradeoff: balancing time interval  throughput latency General Cases

19. 19 Contents  Load balancing Purpose Definition General cases Active load balancing Passive load balancing Special cases Execution of a new task CPU’s shut down or intentionally being IDLE Limitation Load Balance in Linux 2.6.32

20. 20 Special Cases  Execution of a new task When a new task is created in one core, kernel checks the core’s load whether it is reasonable to handle a new task If the load is unacceptable, current task of the core is migrated to the idlest core’s run-queue and rescheduled And a new task is executed in the core (not the idlest core)  CPU’s shut down or intentionally being IDLE When one core should be shut down or intentionally be IDLE, such as in POWER_SAVING_LOAD_BALANCE All tasks in its run-queue are migrated to other cores Actually, this case is just a task migration Load Balance in Linux 2.6.32

21. 21 Contents  Load balancing Purpose Definition General cases Active load balancing Passive load balancing Special cases Execution of a new task CPU’s shut down or intentionally being IDLE Limitation Load Balance in Linux 2.6.32

22. 22 Limitation  Global Fairness Global Fairness 는 여러 개의 CPU 로 이루어진 SMP 에서 모 든 task 가 자신의 weight 에 비례해서 run-time 을 보장받는 정 도를 의미한다. SMP 환경에서 Run queue 가 CPU 에 하나씩 있고, Load Balance 는 각 Run queue 의 load(sum of weight) 만을 고려해 서 task 를 옮기므로 task 가 자신의 weight 에 비례한 시간을 못 받는 경우가 생긴다. Example) Dual-core CPU 에 서로 같은 weight 를 갖는 task1, 2, 3 가 있을 때 CPU1 의 Run-queue 에는 task1 이 있고, CPU2 의 Run-queue 에는 task2, task3 이 들어간다. 이 경우 load balance 가 잘 일어나지 않으므로 서로 같은 weight 를 갖고 있음에도 같 은 run-time 을 보장 받지 못한다. Load Balance in Linux 2.6.32

23. 23 End  Q & A? CFS in Linux 2.6.37