Parallel and Distributed Computing Chapter 7
- 2. OUTLINE Scheduling in Multiprocessing Load Balancing
- 3. MULTIPLE PROCESSOR SCHEDULING Multiprocessors is a system that has more than one processor With multiprocessors load sharing becomes feasible but it makes scheduling more complex. no such policy that can declared best scheduling algorithm for single processor similarly there is no best scheduling solution for multiprocessor systems Best scheduling algorithm depends on the different scenario.
- 4. MULTIPLE PROCESSOR SCHEDULING The scheduling process of a multi-processor is more complex than that of a single processor system because of the following reasons. • Load balancing is a problem since more than one processors are present. • Processes executing simultaneously may require access to shared data. • Cache affinity should be considered in scheduling.
- 5. Characteristics of a Good Scheduling Algorithm: must be flexible enough to process migration decisions when there is a change in the system load A good scheduling algorithm has the property of being scalable. It is flexible for scaling when the number of nodes increases in a system algorithm should not be affected by the failure of one or more nodes of the system
- 6. Categories of Multiprocessors Loosely coupled / Distributed multiprocessors: processors are independent of each other. Each processor has its own memory and I/O Channels. Functionally specialized processor: there is a master/main processor which is general purpose process. It controls other processors in the system and provides services to them Tightly Couple Multiprocessors: all processors shares common memory. Also termed as homogeneous
- 7. Techniques for multiprocessor Scheduling Multi processor scheduling can be done in two ways Asymmetric Scheduling Symmetric Scheduling
- 8. Symmetric multiprocessor Processors are self schedule Each processor check ready queue and select process to execute. If one processor goes down rest of the system keeps working. There are two variants of symmetric multiprocessor Symmetrical Scheduling with global queues. Symmetrical Scheduling with private global queues.
- 9. Symmetrical Scheduling with global queues: if processes are executing in common / global queue scheduler for each processor checks this global queue and select process to execute.
- 10. Symmetrical Scheduling with private queues if processors in the system have their own private ready queue, then processor will check their own private ready queue to select process. Mostly symmetric multiprocessing is used most modern operating system i.e. LINUX, WINDOWS, MAC OS.
- 11. Asymmetric Multiprocessor one processor (master processor) handles all operations and other system activities master processor runs operating system code and other slave processors only executed user code. If master processor fails, whole system comes to a halt, if slave goes down, rest of the system keeps working.
- 12. Processor Affinity A process has an affinity ()تعلق for the processor on which it is currently running. When a process runs on a specific processor, It has effects on the cache memory. The data most recently accessed by the process populate the cache for the processor
- 13. Processor Affinity when process migrates to another processor, now contents of cache memory becomes invalidated for the first processor. Cache for second processor must be repopulated Because of high cost of invalidating and repopulating caches, mostly SMP try to avoid migrating processes from one processor to another. This is known as processor affinity.
- 15. Soft AFFINITY When an operating system has a policy of keeping a process running on the same processor but not guaranteeing it will do so, this situation is called soft affinity.
- 16. Hard AFFINITY ◦When an operating system has a policy of attempting to keep a process running on the same processor with guarantee that it will do so, we have a situation known as hard affinity. ◦Some Linux systems implement soft affinity and provide system calls like sched_setaffinity() that also support hard affinity.
- 17. Assignment a processor can access its own local memory faster than non-local memory (memory local to another processor or memory shared between processors (NUMA) is a computer memory design used in multiprocessing, where the memory access time depends on the memory location relative to the processor Write a short note on Purpose and need of NUMA Architecture
- 18. Load BALANCING balancing of load among Processors of distributed computing. Excess load of one CPU is distributed to other that have less load according to defined limits i.e. neither it gets overloaded nor idle. Load balancing can help to maximize the performance and throughput which is ultimate goal of distributed computing.
- 20. Static load balancing While distributing load current state of the system is not taken into account. These algorithms are simpler Two types 1. Deterministic 2. Probabilistic
- 21. Static load balancing 1. Deterministic: Properties of nodes and processes are taken into account for the allocation of processes to nodes E.g. if tasks are independent, execution time is known. Difficult to achieve better results
- 22. Static load balancing 2. Probabilistic: Algorithm, statistical attributes of the CPU are taken into account to make process placement rules It also doesn’t give better performance. *Statistical attributes: idle time, busy time
- 23. Dynamic load balancing While distributing load, current state of the system is taken into account. Dynamically redistributing workloads away from overloaded nodes toward underloaded nodes. Difficult to design however better performance.
- 24. Dynamic load balancing Types of dynamic load balancing 1. Centralized 2. Distributed
- 25. Dynamic load balancing 1. Centralized: task of handling requests for process scheduling is carried out by centralized server node. Benefit is performance, as all information is held at a single node, but suffers from reliability problem Another problem with increasing number of requests
- 26. Dynamic load balancing 2. Distributed: task of handling requests for process scheduling is carried out by individual node of the system. Unlike centralized, there is no need to hold information. Hence speed is fast.
- 27. Dynamic load balancing Two types of Distributed load balancing ◦A. Cooperative: Scheduling is done with the cooperation of nodes in a system. However drawback is its complexity overhead ◦B. Non-Cooperative: Scheduling is done by individual node in the system as they act as autonomous entities.
- 28. Issues in load balancing Load Estimation Policies: Determination of load of a node Process transfer Policies: decides for the execution of process, local or remote State information Exchange: exchange of information among nodes Location Policy: selection of nodes for migration of the process Priority Assignment: priority making algorithm Migration limit Policy: Determines the limit value for migration of process.