Real-Time/Embedded Operating Systems & Resource Management

(1)

Real-Time/Embedded Operating Systems & Resource Management

Tei-Wei Kuo, Ph.D.

ktw@csie.ntu.edu.tw

Dept. of Computer Science &

Information Engineering National Taiwan University Taipei, Taiwan, ROC

Overview

The Purposes of Operating Systems

Convenience

Efficiency

Characteristics of Many Real-Time/

Embedded Applications

More specific in their applications.

More drastic for their failures.

Overview

A Typical Control System Example

Rates -sensors & actuators, peripheral, control program

Phases - takeoff, cruise, and landing, etc.

sensors

actuators environment controlled

process

Task Executions

Clock

Display operator

(3)

Overview

Potential Timing Hazards:

Loop

…...

Sensor();

……..

computation……

……..

t = time();

SleepTime := ReadyTime + PERIOD - t;

ReadyTime = ReadyTime + PERIOD;

Sleep(SleepTime);

EndLoop;

Loop Size?

Time Elapsed Here?

Timer Granularity?

Real Sleep Time?

???Multiprogramming???

Overview

General Concerns:

Could I verify the performance of my system?

How to avoid timing hazards?

Is there any good way in scheduling processes or allocating resources?

Æ Understand your operating system and hardware, and use resources intelligently!

(4)

A General Architecture

Bottom Half Top Half processes User

Space

OS

hardware

Timer expires to

• Expire the running process’s time quota

• Keep the accounting info for each process

System calls such as I/O requests which may cause the releasing CPU of a process!

Interrupts for Services

(6)

A General Architecture

2-Step Interrupt Services

Immediate Interrupt Service

Interrupt priorities > process priorities

Time: Completion of higher priority ISR, context switch, disabling of certain interrupts, starting of the right ISR (urgent/low-level work, set events)

Scheduled Interrupt Service

Usually done by preemptable threads

Remark: Reducing of non-preemptable code, Priority Tracking/Inheritance (LynxOS), etc.

ISR

I

Interrupt/ISR Latency

SecheduledService

IST Latency

A General Architecture

Scheduler

A central part in the kernel

The scheduler is usually driven by a clock interrupt periodically, except when voluntary context switches occur – thread quantum?

Timer Resolution

Tick size vs Interrupt Frequency

10ms? 1ms? 1us? 1ns?

Fine-Grained hardware clock

(7)

A General Architecture

Memory Management

No protection for many embedded systems

Memory-locking to avoid paging

Process Synchronization

Sources of Priority Inversion

Nonpreemptible code

Critical sections

A limited number of priority levels, etc.

Scheduling Strategies &

System Analysis

Why a process in my application does not meet its deadline?

Factors:

Impacts from the executions of higher- priority processes – preemption cost

Lengthy execution time of the process

Blocking time from lower-priority processes – priority inversion

Scheduling Strategies &

System Analysis

Possible Questions:

How do I assign priorities to processes?

How are my processes scheduled by the OS?

How long is the blocking time/non-

preemptable critical sections (from lower- priority processes or interrupts)?

Æ Understand your schedulers

Fixed-Priorities or Dynamic Priorities

Preemptive or Non-Preemptive Scheduling

(9)

Scheduling Strategies & System Analysis Scheduling Strategy

Major Components of a Scheduler

Priority Assignment Policy

The number of priority levels, e.g., 256?

Aging Effects?

Priority-Driven Scheduling Mechanism

Priority Queue

Thread Quantum?

Preemption Lock – Disabling of Preemption

etc.

Rate Monotonic Scheduling Algorithm

Assumptions:

all periodic fixed-priority processes

relative deadline = period

independent process - no non-preemptable resources

Rate Monotonic (RM) Scheduling Algorithm

RM priority assignment: priority ~ 1/period.

preemptive priority-driven scheduling.

Example: T1 (p1=4, c1=2) and T2 (p2=5, c1=1)

T1 T2 T1 T2 Time

0 1 2 3 4 5 6 7 8

(10)

Rate Monotonic Scheduling Algorithm

Critical Instant ¹

An instant at which a request of the process have the largest completion/response time.

An instance at which the process is requested simultaneously with requests of all higher priority processes

Usages

Worst-case analysis

Fully utilization of the processor power

Example: T1 (p1=4, c1=2) and T2 (p2=5, c1=1)

1 Liu and Layland, “Scheduling Algorithms for multiprogramming in a hard real-time Environment,” JACM, vol. 20, no. 1, January 1973, pp. 46-61.

T1 T2 T1 T2 Time

0 1 2 3 4 5 6 7 8

T2 T2

Rate Monotonic Scheduling Algorithm

Schedulability Test:

A sufficient but not necessary condition

Achievable utilization factorα

of a scheduling policy P -> any process set with total utilization factor no more than α is schedulable.

Given n processes, α =

Stability:

Let processes be sorted in RM order. The ith process is schedulable if

An optimal fixed priority scheduling algorithm

c p

i i

∑

( )

n 2^1/ⁿ −1

( )

c p^j i

j j

i₌ ≤ i −

∑

1

21^/ 1

(11)

Schedulability Tests – A Sufficient Condition

Theorem 0 [Liu&Layland 73]: Aset of n periodic processes is schedulable if the total utilization factor of the process set is no larger than

Theorem 1 [Kuo, et al. 00]: Suppose that T_i-1 is schedulable. Let k be the number of roots in T_i.

If the total utilization factor of T_iis no larger than

then T_iis schedulable.

) 1 2

( ¹ⁿ − n

) 1 2

( ¹^k − k

15 60

3

20

5

Schedulability Tests – Effects of Interrupts

Task Γ1 : C1 =20ms, P1 =100ms, U1=0.2 Task Γ2 : C2 =40ms, P2 =150ms, U2=0.267 Interrupt : Cint=60ms, Pint=200ms, Uint=0.3 Task Γ3 :C3 =20ms, P3 =350ms, U3=0.057

Γ1 Γ2

Γ3 Int

10 100 200 300

Exec with RM priority

Γ1 Γ2

Γ3 Int

10 100 200 300

Exec with an Interrupt priority

The last task was not affected!

(12)

Schedulability Tests – Effects of Priority Mapping

Ready Queue

…

Priority-decreasing

0-3 4-7 8-11

proc[i] proc[j]

proc[k]

allproc

zombproc

freeproc

proc[j] …

proc[m] …

proc[n] …

exit() wait()

Let processes be sorted in RM order. The ith process is schedulable if

( )

(c ) ^/

p

c B

p i

j j

i i

i j

i + + i

≤ −

=

∑

−1

1 21 1

Rate Monotonic Analysis – A Sufficient & Necessary Condition

Rate Monotonic Analysis (RMA) ²

Basic Idea:

Before time t after the critical instance of process τ_i, a high priority process τ_jmay request amount of

computation time.

Formula:

A sufficient and necessary condition and many extensions...

2 Sha, “An Intorduction to Rate Monotonic Analysis,” tutorial notes, SEI, CMU, 1992

Time

c t

j p

j













t

deadline of τ_i

t pj













0

for some t in

 

{kp jj| =1,..., ;i k=1,..., pi/pj }

( ) i

i j

j j

i t d

p c t t

W ≤ ≤











= ∑ = 1 

(13)

Rate Monotonic Analysis – A Sufficient & Necessary Condition

A RMA Example:

T1(20,100), T2(30,150), T3(80, 210), T4(100,400)

T1

c1 <= 100

T2

c1 + c2 <= 100 or

2c1 + c2 <= 150

T3

c1 + c2 + c3 <= 100 or

2c1 + c2 + c3 <= 150 or

2c1 + 2c2 + c3 <= 200 or

3c1 + 2c2 + c3 <= 210

T4

c1 + c2 + c3 + c4 <= 100 or

2c1 + c2 + c3 + c4 <= 150 or

.... Time

W3(t)

50 100 150 200 130

150 170 190 210

Rate Monotonic Scheduling Algorithm

RM was chosen by

Space Station Freedom Project

FAA Advanced Automation System (AAS)

RM influenced

the specs of IEEE Futurebus+

RMA is widely used for off-line analysis of time-critical systems.

(14)

Earliest Deadline First Scheduling Algorithm

Assumptions (similar to RM):

all periodic dynamic-priority processes

relative deadline = period

independent process - no non-preemptable resources

Earliest Deadline First (EDF) Scheduling Algorithm:

EDF priority assignment: priority ~ absolute deadline.

i.e., arrival time t + relative deadline d.

preemptive priority-driven scheduling

Example: T1(c1=1, p1=2), T2(c2=2, p2=7)

T1 T2 Time

0 1 2 3 4 5 6 7 8

T1 T2 T1 T1 T2

Earliest Deadline First Scheduling Algorithm

Schedulability Test:

A sufficient and necessary condition

Any process set is schedulable by EDF iff

EDF is optimal for any independent process scheduling algorithms

However, its implementation has

considerable overheads on OS’s with a fixed- priority scheduler and is bad for (transiently) overloaded systems.

c p

j

j j

i₌ ≤

∑

1 1

(15)

Synchronization & IPC

Why Inter-Process Communication (IPC)?

Exchanging of Data and Control Information!

Why Process Synchronization?

Protect critical sections!

Ensure the order of executions!

(16)

Synchronization & IPC

Common Facility for IPC?

Shared Memory

Message Transmission

Common Facility for Synchronization?

Semaphores

Signals

Synchronization & IPC

Shared Memory

Characteristics: High

bandwidth and low latency, but very primitive!

Needs: Ways to

synchronize its access to avoid racing conditions!

Potential deadlocks/

livelocks/blocking problems

e.g., semaphores, bakery algorithm, etc.

flag[i]=TRUE;

while flag[j]

;

…

flag[i]=FALSE;

key=TRUE;

for(swap(lock,key); key==TRUE; ) swap(lock,key) ;

…

lock=FALSE;

Example 1:

Example 2: (initially, lock=FALSE)

Deadlock?

Livelocks?

(17)

Synchronization & IPC

Message Transmission

Characteristics: Simple & clean

interface with various extensions: m:m communication, multi-machines

Needs: A priority-based message transmission/notification/processing mechanism

e.g., message priority, non-blocking library functions, notification of msg arrivals, servicing order of msgs (with respect to other threads in the system), etc.

A B

OS

Synchronization & IPC

General Concerns:

How to enforce mutual exclusion?

We should not rely on OS scheduling to avoid race conditions!

How to process critical messages first?

There is a tight coupling between

message processing and OS scheduling!

After all, we want to manage the priority inversion problem!

How to let critical jobs be done with minimized interferences from less important jobs!

(18)

Synchronization & IPC

Solutions??

Synchronization Methods

Priority-Driven Resource Scheduling Support!

Popular Approaches

Semaphore-Based Synchronization?!

Signals are too slow.

Priority Inheritance?!

Ceiling?

Synchronization & IPC – Signals?

The Design of Signal Mechanisms

Inform processes/threads of the occurrences of exceptions or events

Posting of a signal to a process

An appropriate signal is added to the set of pending signals for the process.

Delivering within the context of the receiver

if (sig = CURSIG(p)) postsig(sig)

Signal handlers: user-mode routines

A B

OS

…

(19)

Synchronization & IPC – Signals

Questions?

How fast can a signal be delivered?

Complicated actions done by OS

The length of a signal handler?

Signal handlers are executed prior to returning control to the user code.

Real-Time Support

Application-defined signals

Queuing of signals while being blocked.

Signals are delivered in the priority order.

etc

Synchronization & IPC – Can we manage the priority inversion problem?

What are the sources of priority inversion?

Synchronization and mutual exclusion

Nonpreemptable regions of code

FIFO of any other non-priority-based queues

Interrupts

A limited number of priorities

(20)

Synchronization & IPC – Can we manage the priority inversion problem?

Popular Synchronization Methods

Nonpreemptable Critical Sections

Highest Locker’s Priority

Priority Inheritance

Priority Ceiling

1. Nonpreemptible Critical Section Critical sections are executed at an

“infinitely” high priority!

Synchronization Protocols

τ_L τ_M

τ_H

Note that τ_H&τ_Mhave no intention to enter a critical section!

Time

(21)

Synchronization Protocols

2. Highest Locker’s Priority Protocol

Execute critical section at the priority of the highest-priority task that may lock the semaphore(/resource); higher-priority tasks may preempt the critical section.

τ_L τ_M τ_H τ_vH

Note that τ_vHis no longer blocked by τ_L

Time

3. Basic Inheritance Protocol (BIP) [Sha87]

Execute the critical section at the priority of the highest-priority task being blocked; higher-priority tasks may preempt the critical section.

τ_L S1

τ_M τ_H τ_vH

S1

S2

t

blocked blocked

Time

- Note that τ_Mis no longer blocked until necessary.

- However, system may be deadlocked or have chained blocking!

Synchronization Protocols

S2 S1 & S2

(22)

4. Priority Ceiling Protocol [Sha87]

BIP + a “Priority Ceiling” rule about when to grant lock requests (see [Sha 87, 90] )

τ_L S1

τ_M τ_H τ_vH

S1

S2 S1

blocked Time

- No deadlock & chained blocking at the cost of reducing the concurrency level of the system.

- Blocked-at-most-once.

Synchronization Protocols

blocked

S1

Summary of Synchronization Methods

Yes Yes

Priority Ceiling

No Bounded Priority Inheritance

Yes Yes

Highest Locker’s Priority

Yes Yes

Non-preemptible Critical Section

Deadlock Avoidance Blocked

at Once

(23)

Real-Time/Embedded Operating Systems & Resource Management