1. Chapter 19: Real-Time Systems

2. Chapter 19: Real-Time Systems
Overview and Introduction
System Characteristics
Features of Real-Time Systems
Implementing Real-Time Operating Systems
Real-Time CPU Scheduling

3. Objectives To explain the timing requirements of real-time systems
To distinguish between hard and soft real-time systems
To discuss the defining characteristics of real-time systems
To describe scheduling algorithms for hard real-time systems

4. Overview of Real-Time Systems
The differences between real time computing systems and general purpose computing systems are very profound.
We will examine many of these differences in the upcoming slides.
A real time computing system is one that requires that correct results be produced within specified deadline periods. (Deadline is a key word)
Results produced after a specific time period has elapsed may well be of absolutely no value and may mean loss of life or an aircraft crash; yet other failures might not be quite as disastrous if real time system is not quite as responsive. .
Real time:
Run on wide range of computer hardware;
Used in many different kinds of applications.
Some real time systems are embedded in aircraft instrumentation:
your microwave, cell phone, cruise control, and hosts of other applications.
Are often part of a larger system;
Oftentimes their presence is not obvious to a user.

5. Definitions
A real-time system requires results produced within a specified deadline period.
A hard real time system has stringent requirements, guaranteeing that critical real-time tasks be completed within their deadlines.
E.g. Safety-critical systems; health systems; etc.
A soft real-time system is less restrictive and guarantees simply that a critical real-time task will receive priority over other tasks;
Further, these tasks retain priority over other tasks until the tasks complete.
An embedded system is a computing device that is part of a larger system (I.e. automobile, airliner.)
A safety-critical system is a real-time system with catastrophic results in case of failure.
Again, a hard real-time system guarantees that real-time tasks be completed within their required deadlines, while a soft real-time system provides priority of real-time tasks over non real-time tasks.

6. System Characteristics
Will look at both soft and hard real-time operating systems..
Real time systems typically exhibit the following characteristics:
Single purpose
Small size
Inexpensively mass-produced
Specific timing requirements
And, so many other rather unique features spring from these characteristics.
These are the big four defining characteristics…
Let’s look at those…

7. Characteristics - 1 Single Purpose:
Single purpose is typical: controlling anti-lock brakes; toaster, cell phone.
This makes the operating system simple too, as many characteristics integral to general – purpose operating systems are not available or needed.
Size:
Often found in severely cramped space – but sufficient for operations.
Examples: wrist watches, cell phones, toys
Thus, CPU processing power is minimal
Amount of primary memory is also minimal.
Architectures: Compare 32 / 64-bit architectures with 8/16 bit processors.
Architecture: Compare several gigs of memory with < 1 MB memory.

8. Characteristics - 2 Cost
Typically mass produced: as in microwave ovens, thermostats.
Thus, real time microprocessors are often inexpensive
Organization of real time systems are designed to minimize cost:
To eliminate bus architectures, the physical organization for embedded controllers are often organized as a system-on-a-chip, which has all necessary interconnections.
Chip includes memory, cache, a MMU (for possible address translation), any peripheral ports necessary - all in a single integrated circuit.
Such organization is typically much less expensive than typical bus-oriented architectures.

9. Bus-Oriented System

10. Characteristics - 3 Timing:
This is the feature that impacts almost everything else and makes real time systems what they really are!
Both ‘hard’ and ‘soft’ real time systems have timing requirements.
So, we will need to develop real time scheduling algorithms that provide priority to the highest scheduled processes.
Schedulers absolutely must ensure that the priority of real time tasks does not degrade over time.
So, we must minimize the response time to interrupts – recognize the interrupt, save context, transfer control to the handler, etc… as one can easily imagine.

11. Features of Real-Time Kernels
Most real-time systems do not provide features found in desktop systems.
Simply not needed.
Do not need (in general) support for
A variety of peripheral devices – graphical displays, CD, DVD drives…
Protection and security mechanism
Support for multiple users…
Note: Windows XP has > 40,000,000 lines of code; a typical real-time operating system usually is written in thousands of source code lines!
Reasons include
Real-time systems are typically single-purpose.
Real-time systems often do not require interfacing with a user.
General-purpose features often require more substantial hardware than that found in a RT system.

12. Memory Mapping Schemes
(1) Real-addressing mode where programs generate actual addresses.
But there is no memory protection between processes.
We might also need to specify exactly where in memory the program is to be loaded.
But the speed is very difficult to beat!
No time spent in address translations.
Very commonly found in real time systems having hard, real-time constraints.
Some real time operating systems running on microprocessors containing a MMU, actually disable the MMU to gain performance benefit in referencing physical addresses directly.
(2) Relocation register mode.
In this scheme, we have a relocation register is set to the process’ load point.
Physical addresses are added to contents of the relation register formed by adding logical (L) to R (relocatable) to form P (Physical) addresses..
But here again, there is no protection between processes.
(3) Implementing full virtual memory.
But here we may have page tables and translation look-aside buffers.
While this strategy does indeed provide memory protection, it is costly…

13. Address Translation
Diagram of the previous three memory access techniques.

14. Implementing Real-Time Operating Systems
In general, real-time operating systems must provide the following features:
(1) Preemptive, priority-based scheduling
(2) Preemptive kernels
(3) Minimal latency
Let’s look at these three important characteristics.

15. 1. Priority-Based Scheduling
A RTOS must respond immediately to a real time process as soon as that process requires the CPU.
So, there must be a scheduler to support a priority-based algorithm with preemption.
In priority-based scheduling, processes are assigned a priority based on importance.
More important processes are assigned higher priorities.
So, if the scheduler supports preemption, a process running on a CPU can be preempted if a higher-priority process arrives.
Recall: Windows XP has 32 priority levels, with the highest levels having priority values 16 to 31 used for RT processes..

16. 2. Preemptive Kernels
Problem with non-preemptive kernels is process doesn’t have to give up the CPU. This can be disastrous in a real time system.
Preemptive kernels allow the preemption of a task running in kernel mode!
But designing preemptive kernels is complex
Many common modern-day applications (spreadsheets, browsers, ..) simply do not require preemptive kernels.
So why have the complexity of such kernels?
Thus most commercial desktop OSs are not preemptive, such as XP.
So, non-preemptive kernels are not acceptable for hard RT systems.
We need preemption!

17. 2. Preemptive Kernels - 2 One approach
to insert preemption points within long duration system calls, so when a process is preempted, and, if a context-switch takes place, when the process resumes after the high priority process runs, it may do so at the point of preemption!
Important to note that preemption points only occur at carefully architected points in the kernel, where kernel structures are not undergoing any kind of modification.
Second approach for supporting preemptive kernels is
to implement synchronization mechanisms which protect kernel data structures from modification from a high-priority process.

18. 3. Minimizing Latency
Event latency is the amount of time from when an event occurs to when it is serviced.
Here, we consider the event-driven nature of a RT system, and we recognize that the application is usually waiting for an event. .
The system must respond and service the event as quickly as possible!

19. Minimizing Latency – continued
But all events are not created equal.
Different events have different latency times.
Some have a very short latency time; others longer.
Example: an embedded system controlling something like a toaster or some kind of radar, might tolerate a latency of several seconds.
In RT systems we’ve got two types of latencies to consider:
Interrupt latency, and
Dispatch latency.

20. Interrupt Latency
Interrupt Latency: refers to the period of time from the arrival of an interrupt at the CPU to the start of the servicing routine. (see figure below)
Getting an interrupt, the CPU must finish the current instruction, determine type of interrupt, save the state of the current process (context switch likely) and then jump to the interrupt service routine.
We clearly need to minimize interrupt latency and service the interrupt immediately.

21. Dispatch Latency
Dispatch latency is the amount of time required for the scheduler to stop one process and start another.
RTOSs must minimize this latency, and the most effective approach for keeping dispatch latency at a minimum is via preemptive kernels.
See figure. The conflict phase has two parts:
Preemption of any process running in the kernel, and
Release by low-priority process resources needed by a high-priority process.
(e.g. dispatch latency w/preemption disabled in Solaris is > 100 msec; with preemption enabled, preemption is reduced to < 1 msec!

22. Real-Time CPU Scheduling
We must change gears to address the reality that – so far – we are only ensuring that real time systems provide priority processing for critical processes.
But hard real-time systems need much stronger guarantees!
Tasks MUST be serviced by deadline w/knowledge that missing deadline = no service at all!
So let’s consider scheduling considerations for hard real-time systems.

23. Real-Time CPU Scheduling - more
To understand what’s going on, let’s consider the following assumptions and definitions.
First, we will assume that RT processes are periodic.
This means that they need the CPU at constant intervals.
Each of these periods processes a fixed processing time, t, once the CPU acts on it,
Each has a deadline, d, when it must be serviced by the CPU, and
Each has a period, p
The relationship of the processing time t, the deadline, and the period can be expressed as
0 <= t <= d <= p.
The rate of a periodic task is 1/p.
See next figure to see these relationships. .

24. Real-Time CPU Scheduling
Periodic processes require the CPU at specified intervals (periods)
p is the duration of the period
d is the deadline by when the process must be serviced
t is the processing time
Clearly, we need the processing time t to be less than the deadline (be completed before deadline expires).
Let’s look at scheduling algorithms that address deadline requirements of
hard, real-time systems.

25. Preface to Scheduling Algorithms
We will look at the
1. Rate Monotonic Scheduling Algorithm, and
2. Earliest Deadline First Scheduling.
The rate-monotonic scheduling algorithm schedules these periodic tasks using a static priority policy with preemption.
If we are running a lower priority process when a higher priority process needs to be run, the scheduler will preempt the lower priority procedure.
When a periodic task enters the system, it is assigned a priority number based on its period.
Shorter the period, higher the priority, and vice versa.
Gives higher priority to the process that needs the CPU more often.
Procedure also assumes processing time of a periodic process is the same for each CPU burst!
So every time a process acquires the CPU, duration of its CPU burst is the same.
As it turns out using the simulation in the book, it seems that we ‘may’ be able to assert that we can schedule real time tasks in such a way that using this algorithm may be used to meet real time task deadlines and still have the CPU with available cycles…
We shall see….
Now let’s consider two cases: We will assign P2 a higher priority than P1 and then P1 a higher priority than P2.

26. Scheduling of tasks when P2 has a higher priority than P1
Without the rate-monotonic scheduler:
We will first assume that P2 has a higher priority than P1.
We see the execution scenario below.
P2 starts executing first and completes at time 35. Then P1 starts.
It completes its burst at time 55 but the first deadline for P1 was at 50, so
the scheduler causes P1 to miss its deadline!
Now, suppose we continue to use the rate-monotonic scheduling approach where we
assign P1 a higher priority than P2, (since the period of P1 is shorter than P2).

27. Rate Monotonic Scheduling
Continuing: (P1 period shorter than P2…).
P1 starts and completes its CPU burst at time 20. Meets its first deadline.
P2 starts and running until time 50 – but at this time, it is preempted by P1, even though it only needed 5 msec to finish its first CPU burst.
P1 finishes its CPU burst at time 70 (meets its deadline); scheduler resumes P2, which completes its burst at time 75, also meeting its first deadline.
System is then idle until time 100 when P1 is scheduled again.
This rate monotonic scheduling is considered optimal in the sense that if a set of processes cannot be scheduled by this algorithm, it cannot be scheduled by any other algorithm that assigns static priorities.
Your book then goes into another example and arrives at the important conclusion: the rate-monotonic scheduling cannot guarantee processes can be scheduled so that they will always meet their deadlines.
So, enter the Earliest Deadline First Scheduler…. (EDFS)

28. Earliest Deadline First Scheduling
This scheduling algorithm dynamically assigns priorities according deadline.
Earlier the deadline, the higher the priority; later deadline?, lower priority.
In this algorithm, when a process becomes runnable, it must announce its deadline requirements to the system.
Priorities may have to be dynamically adjusted to reflect deadlines of newly runnable processes.
This differs from rate-monotonic scheduling, where priorities are fixed.

29. The EDF scheduling algorithm does not require that processes by periodic nor must a process require a constant amount of CPU time per burst.
The only requirement is that a process announce its deadline to the scheduler when it becomes runnable.
The attraction of the EDF scheduling is that it is theoretically optimal; it can schedule processes so that each process can meet its deadline requirements and CPU utilization will be 100 percent.
In practice, as it turns out, it is impossible to achieve this level of CPU utilization due to the cost of context switching between processes and interrupt handling.

30. End of Chapter 19