Multiprocessor and Real-Time Scheduling

0
0
2756 days ago, 786 views
PowerPoint PPT Presentation
B.Ramamurthy. 2. Presentation. At the point when a PC framework contains more than a solitary processor, a few new issues are brought into the configuration of booking capacities. We will inspect these issues and the points of interest of planning calculations for firmly coupled multi-processor frameworks.. B.Ramamurthy. 3.

Presentation Transcript

Slide 1

Multiprocessor and Real-Time Scheduling Bina Ramamurthy Chapter 10 CSE421 B.Ramamurthy

Slide 2

Introduction When a PC framework contains more than a solitary processor, a few new issues are brought into the outline of booking capacities. We will look at these issues and the points of interest of planning calculations for firmly coupled multi-processor frameworks. B.Ramamurthy

Slide 3

Topics for talk Granularity of calculation Design issues Assignment of procedures to processors Multiprogramming on individual processors Actual dispatching of a procedure Real-time frameworks Characteristics Real-time booking Rate-monotonic planning (RMS) B.Ramamurthy

Slide 4

Granularity The fundamental motivation behind having various processors is to realize____________________? Applications display parallelism at different levels. There are no less than five classifications of parallelism that contrast in the level of granularity. Fine grain parallelism: Parallelism characteristic in a solitary inst. stream. normally does not include the OS but rather done at accumulation organize. High information reliance ==> high recurrence of sync. reqd. B.Ramamurthy

Slide 5

Granularity Medium grain : Potential parallelism of an application can be executed by various strings in a solitary procedure. High level of coordination among the strings. Generally developer determines the parallelism in the outline. Coarse grain : Multiprocessing of simultaneous procedures in a multiprogramming situation :Example : Prj2. Exceptionally coarse grain : When the synchronization required is not all that continuous the procedures can be appropriated among system hubs to frame a solitary processing condition. Case: CORBA standard for dispersed framework Independent parallelism : Multiple irrelevant procedures. B.Ramamurthy

Slide 6

Design Issues Assignment of procedures to processors. Utilization of multiprogramming on individual processors. Genuine dispatching of a procedure. B.Ramamurthy

Slide 7

Process/processor Assignment Master/slave task : Kernel works dependably keep running on a specific processor. Different processors execute client forms. Advantage: Resource struggle determination disentangled since single processor has control. Issue: Failure of ace processor? Ace processor does the planning ==> bottleneck. Peer task : OS can execute on any processor. Every processor does its own particular booking from the pool of accessible procedures. This is like Solaris or NT symmetric multiprocessing (SMP). B.Ramamurthy

Slide 8

Multiprogramming at every processor Completion time and other application-related execution measurements are a great deal more imperative than processor usage in multi-processor condition. For instance, a multi-strung application may require every one of its strings be allocated to various processors for good execution. Static or element designation of procedures. B.Ramamurthy

Slide 9

Process dispatching After task, choosing who is chosen from among the pool of holding up procedures - handle dispatching. Single processor multiprogramming systems might be counter-beneficial here. Needs and process history may not be adequate. B.Ramamurthy

Slide 10

Process booking Single line of procedures or if various need is utilized, different need lines, all sustaining into a typical pool of processors. Multi-server lining model: various line/single line , different server framework. Concentrate the framework for 1, 2, and n processors. Surmising: Specific booking strategy does not have much impact as the processor number increments. Conclusion: Use FCFS with need levels. B.Ramamurthy

Slide 11

Thread planning An application can be actualized as an arrangement of strings that coordinate and execute simultaneously in a similar address space. Criteria: When related strings keep running in parallel perf. moves forward. Stack sharing : pool of strings, pool of processors. Posse booking : Bunch of related strings planned together. Committed processor task : Each program gets the same number of processors as there are parallel strings. Dynamic booking : More like request planning. B.Ramamurthy

Slide 12

Real-time frameworks Real-time registering is an imperative rising control in CS and CE. Control of lab analyses, apply autonomy, handle control, media transmission and so forth. It is a sort of figuring where accuracy of the calculation depends on the intelligent outcomes as well as on the time at which the outcomes are delivered. Hard constant frameworks: Must meet due date. Ex: Space carry meet with other space station. Delicate continuous framework: Deadlines are there however not required. Results are disposed of if the due date is not met. B.Ramamurthy

Slide 13

Characteristics of Real-Time (RT) frameworks Determinism Responsiveness User control Reliability Fail-delicate operation B.Ramamurthy

Slide 14

Deterministic Response External occasion and timings direct the demand of administration. OS's reaction relies on upon the speed at which it can react to hinders and on whether the framework has adequate ability to deal with solicitations. Determinism is worried with to what extent the OS delays before recognizing an intrude. In non-RT this deferral might be in the request of 10's and 100's of millisecs, while in a RT it might have an upper-bound of few microsec to 1millisec. B.Ramamurthy

Slide 15

RT .. Responsiveness is the ideal opportunity for overhauling the hinder once it has been recognized. Includes: Time to exchange control, (and setting switch) and execute the ISR Time to deal with settled intrudes on, the interferes with that ought to be adjusted when executing this ISR. Higher need Interrupts. reaction time = F(responsiveness, determinism) B.Ramamurthy

Slide 16

RT .. Client Control User control : User has a considerably more extensive control in RT-OS than in normal OS. Need Hard or delicate due dates Deadlines Memory Managemenmt: paging or swapping Name the procedures to be occupant in memory Scheduling strategies B.Ramamurthy

Slide 17

RT .. Unwavering quality Reliability : A processor disappointment in a non-RT may bring about diminished level of administration. In any case, in a RT it might be calamitous : life and demise, money related misfortune, gear harm. Bomb delicate operation: Ability of the framework to flop in such a path safeguard however much capacity and information as could be expected. In case of a disappointment, quick identification and redress is imperative. Advise client procedures to rollback. Apply pay. B.Ramamurthy

Slide 18

Requirements of RT Fast setting switch Minimal usefulness (little size) Ability to react to hinders rapidly (Special intrudes on handlers) Multitasking with signs and alerts Special stockpiling to amass information quick Preemptive planning B.Ramamurthy

Slide 19

Requirements of RT (contd.) Priority levels Minimizing hinder debilitates Short-term scheduler ("omni-powerful") Time screen Goal: Complete all hard continuous assignments by dead-line. Finish whatever number delicate continuous undertakings as could be expected under the circumstances by their due date. B.Ramamurthy

Slide 20

Preemption and here and now booking Fig.10.4 B.Ramamurthy

Slide 21

RT planning Static table-driven approach For intermittent assignments. Contribution for investigation comprises of : occasional landing time, execution time, finishing time, needs. Unbendable to element changes. General arrangement: most punctual due date first. B.Ramamurthy

Slide 22

RT booking Static need driven preemptive planning For use with non-RT frameworks: Priority based preemptive planning. Need task in light of continuous requirements. Illustration: Rate monotonic calculation B.Ramamurthy

Slide 23

RT booking Dynamic arranging based planning After the errand touches base before execution starts, a timetable is readied that incorporates the new and additionally the current assignments. In the event that the new one can abandon influencing the current timetables than nothing is amended. Else calendars are updated to suit the new undertaking. Keep in mind that occasionally new assignments might be rejected if due dates can't be met. B.Ramamurthy

Slide 24

RT booking Dynamic best-exertion planning: utilized as a part of most business RTs of today undertakings are aperiodic, no static booking is conceivable some fleeting planning, for example, briefest due date first is utilized. Until the assignment finishes we don't know whether it has met the due date. B.Ramamurthy

Slide 25

Deadline planning Characteristics considered Aperiodic assignments Ready time Starting due date Completion due date Processing time Resource prerequisites Priority B.Ramamurthy

Slide 26

Scheduling undertakings with culmination due date Fig. 10.5, Table 10.2 B.Ramamurthy

Slide 27

Scheduling assignments with beginning due date Fig. 10.6, Table 10.3 B.Ramamurthy

Slide 28

Rate Monotonic Scheduling Fig. 10.7, Fig. 10.8, Table 10.4 B.Ramamurthy

Slide 29

Summary We examined issues and calculations in Multi-processor planning Thread booking Real-time planning B.Ramamurthy

SPONSORS