Real-Time Operating Systems
Lecture for the Embedded Systems Course
CSD, University of Crete (May 23, 2014)
Manolis Marazakis (maraz@ics.forth.gr)
Institute of Computer Science (ICS)
Foundation for Research and Technology – Hellas (FORTH)
Real-Time OS = ?
Real-Time
Control External Events
Synchronous, Asynchronous, Independent
Speed
Fast response, Low overheads
Deterministic
A late answer is a wrong answer
Multi-Tasking
Sequence: One task controlling all events in an infinite loop
Multi-tasking: Task “Waiting” on Event, Task becomes “Ready” upon Event
Priority Scheduling
Preemptive Scheduling on Task Priority
No delay on Context Switch to the next Tick
Equal Priority tasks won’t preempt each other
Scheduling due to synchronous or asynchronous event
Round-Robin Scheduling
All tasks in the same priority share the CPU
Context Switch in each (predefined) time slice period
2 Real-Time Operating Systems
Need for Synchronization ?
Reentrancy & Shared Resources
Co-ordination between tasks
Asynchronous events
Interrupts & Exceptions
Interrupts allow devices to notify the CPU on events
Interrupt Levels
Exceptions
Unplanned event generated by the CPU (e.g. divide-by-0)
Exceptions will generate an “internal” interrupt
ISR (Interrupt Service Routines)
Not a task (has no “task control block”)
Timers
Synchronous events
3 Real-Time Operating Systems
RTOS vs General-Purpose OS
General-purpose OS:
Key criteria: Task throughput, fairness
“average case” optimizations
Real-Time OS:
Deterministic timing behavior
“worst-case” guarantees
Processing must be done within a time constraint or the system will fail.
All delays in the system will be bounded; from the retrieval of stored
data to the time RTOS finishes the request.
4 Real-Time Operating Systems
Can Linux provide real-time guarantees ? (NO)
(usually) Non-preemptive kernel
A system call may sometimes take a “long time” to complete
Coarse timer resolution
Tasks can be released with millisecond precision (1/10/100)
Virtual memory
Unpredictable delay
Variable task priority
Task priority varies with time, to achieve fairness
Batching and possible re-ordering of storage & network
operations (for efficient use of H/W)
High-priority task may queue behind a low-priority task that has
issued a system call
High-priority tasks may have to wait for lower priority tasks to
release resources
5 Real-Time Operating Systems
Can Linux provide real-time guarantees ? (YES)
Provides RT-POSIX interface
Fixed priority real-time scheduling classes
SCHED_FIFO
SCHED_RR
Timesharing in SCHED_OTHER
Preemptive Linux kernel …
Concerns:
• Real-time ?
• Footprint ?
6 Real-Time Operating Systems
Outline
RTOS “theory”
POSIX 1003.1b standard
LynxOS
• Asynchronous I/O
QNX • Semaphores
Wind River VxWorks • Message queues
RTLinux • Queued signals
TimeSys Linux • Scheduling
• Clocks &timers
• Memory Management
7 Real-Time Operating Systems
Three Models of Kernel
Kernel = ?
Process Management
Memory Management
I/O System Management
Three models of “kernel” :
Monolithic
Layered
Microkernel (client-server)
8 Real-Time Operating Systems
Monolithic Kernel
In a monolithic-modularized OS, the functionality is
integrated into a single executable file that is made up of
modules, separate pieces of code reflecting various OS
functionality.
9 Real-Time Operating Systems
Layered Kernel
In the layered design, the OS is divided into hierarchical layers
(0...N), where upper layers are dependent on the functionality
provided by the lower layers (via APIs).
Like the monolithic design, layered OSes are a single large file that
includes device drivers and middleware
10 Real-Time Operating Systems
Microkernel (Client-server) OS
An OS that is stripped down to minimal functionality,
commonly only process and memory management subunits
is called a client-server OS, or a microkernel.
The remaining functionality typical of other kernel algorithms is
abstracted out of the kernel, while device drivers, for instance,
are usually abstracted out of a microkernel entirely
11 Real-Time Operating Systems
Non-preemptive Scheduling
Tasks are given control of the master CPU until they have finished
execution, regardless of the length of time or the importance of
the other tasks that are waiting.
First-Come-First-Serve (FCFS)/ Run-To-Completion
The response time of a FCFS algorithm is typically slower than other algorithms
Shortest Process Next (SPN)/Run-To-Completion
SPN algorithm has faster response times for shorter processes. However, then
the longer processes are penalized by having to wait until all the shorter
processes in the queue have run. In this scenario, starvation can occur to
longer processes if the ready queue is continually filled with shorter processes.
The overhead is higher than that of FCFS, since the calculation and storing of
run times for the processes in the ready queue must occur.
Co-operative: the tasks themselves run until they tell the OS when they
can be context switched (i.e., for I/O, etc.).
This algorithm can be implemented with the FCFS or SPN algorithms, rather
than the run-to-completion scenario, but starvation could still occur with SPN
if shorter processes were designed not to “cooperate,” for example.
12 Real-Time Operating Systems
Preemptive Scheduling
Round Robin/FIFO (First In, First Out) Scheduling
Each process in the FIFO queue is allocated an equal time slice (the duration
each process has to run). An interrupt is generated at the end of each of
these intervals to start the pre-emption process.
Priority (Preemptive) Scheduling
Every process is assigned a priority
The processes with the highest priority always preempt lower priority
processes when they want to run
Problems:
Process starvation, Priority inversion, how to determine the priorities of
various processes, i.e. same priority vs how often the process(es) are run
13 Real-Time Operating Systems
Preemptive Scheduling (cont.)
Rate Monotonic Scheduling (RMS) scheme, in which
tasks are assigned a priority based upon how often
they execute within the system.
EDF (Earliest Deadline First)/Clock Driven Scheduling
The EDF/Clock Driven algorithm schedules priorities to
processes according to three parameters:
frequency (number of times process is run)
deadline (when processes execution needs to be
completed)
duration (time it takes to execute the process)
14 Real-Time Operating Systems
Preemptive Scheduling & RTOS
Tasks with real-time requirements have to be allowed to
preempt other tasks
Preemptive scheduling must be one of the algorithms implemented
within RTOS schedulers
RTOS schedulers also make use of their own array of timers, ultimately based
upon the system clock, to manage and meet their hard deadlines.
Whether an RTOS or a non real-time OS in terms of scheduling,
all will vary in their implemented scheduling schemes. E.g:
vxWorks (Wind River) : priority-based + round-robin
Jbed (Esmertec) : EDF
Linux (Timesys) : priority-based
15 Real-Time Operating Systems
LynxOS (1/2)
Microkernel design
Small kernel footprint : ~28 KB in size
The microkernel provides essential services in scheduling, interrupt
dispatching and synchronization
The other services are provided by kernel lightweight service
modules Kernel Plug-Ins (KPIs)
New KPIs can be added to the microkernel to support I/O, file
systems, TCP/IP, streams and sockets
Can function as a multipurpose UNIX OS
LynxOS offers memory protection through hardware MMUs
Applications make I/O requests to I/O system through system calls
Kernel directs I/O request to the device driver
16 Real-Time Operating Systems
LynxOS (2/2)
Each device driver has an interrupt handler and kernel
thread
The interrupt handler carries the 1st step of interrupt handling
If it does not complete the processing, it sets an asynchronous trap
to the kernel
Later, when kernel can respond to the software interrupt, it
schedules an instance of the kernel thread to complete the
interrupt processing
17 Real-Time Operating Systems
QNX Neutrino (1/2)
Microkernel design – kernel provides essential threads and real-
time services
Microkernel footprint: ~12 KB
Other services are considered as resource managers and can be
added or removed at run-time
SMP RTOS – requires high end, networked SMP machines with
GBs of physical memory
Message passing:
Messages are the basic means of inter-process communication among
all threads
Message-based priority tracking
Messages are delivered at the priority order and the service provider
executes at the priority of the highest priority clients waiting for service
So, if the highest priority task wants to do read some data from file, the file
system resource manager will execute at this task’s priority
18 Real-Time Operating Systems
QNX Neutrino (2/2)
When a service provider thread wants to provide service,
then it creates a channel (for exchanging messages) with its
service identifier for identification
To get a service from a provider, the client thread attaches it to
the provider’s channel
Within the client, this connection is directly mapped to the file
descriptor (so RFS can be sent directly to the file descriptor)
QNX messages are blocking (unlike POSIX standards)
19 Real-Time Operating Systems
VxWorks (1/4)
Microkernel Monolithic Kernel (CoreOS + Wind microkernel)
Provides interfaces specified by RT-POSIX standards in addition to
its own APIs
Shared-memory objects: shared binary and counting semaphores
Standard MMU (as in modern OS)
Provides basic virtual-to-physical memory mapping
Allows to add new mappings and make portions of memory non cacheable
When memory boards are added dynamically, to increase the address space
for inter-process communication
The data is made non cacheable, to ensure cache consistency
20 Real-Time Operating Systems
VxWorks (2/4)
Reduced Context Switch time
Saves only those register windows that are actually in use
(e.g. applicable on SPARC)
When a task’s context is restored, only the relevant register window is
restored
To improve response time, it saves the register windows in a register cache
useful for recurring tasks
21 Real-Time Operating Systems
VxWorks (3/4)
Real-Time Embedded Application
Graphics Multiprocessing Internet
Java Support POSIX Library File System
WindNet Networking
Core OS: Wind Microkernel
22
22 Real-Time Operating Systems
VxWorks (4/4)
23 Real-Time Operating Systems
RTLinux (1/3)
Patch-set for Linux kernel + RT API
Runs Linux kernel as lowest-priority task (pre-emptible)
User Processes
System libraries
Device drivers Linux kernel
I/O Hardware Interrupts
Hardware
24 Real-Time Operating Systems 24
RTLinux (2/3)
Linux is executed in the background
User Processes
Real Time Tasks System libraries
Device drivers Linux kernel
I/O Software Interrupts
Direct
h/w RT-Scheduler
access RTLinux Plug-in
I/O Hardware Interrupts
Hardware
25 Real-Time Operating Systems 25
RTLinux (3/3)
Development flow for RT-application program:
1. Develop RT-application program : Linux kernel module
2. Load RT-Core / RT-Scheduler / RT-FIFOs transit to RTLinux
3. Load RT-application module
Interrupt control H/W
Real-Time Kernel
Real-Time
Linux
Tasks
Real-Time
Linux FIFOs
Processes
26 Real-Time Operating Systems
Linux/RK
A Kernel that provides to applications Timely, Guaranteed, and Enforced access
to System Resources
Allows Applications to specify only their Resource Demands
… leaving the Kernel to satisfy Demands using hidden management schemes
Linux Linux Linux Reservation
Process Process Process Parameters
“T”: Period (1/f)
User-Level “C”: Execution
time within
Kernel period
Resource
“D”: Deadline
Kernel Linux within period
Kernel
Loadable Kernel
Kernel
Module
Hardware
27 Real-Time Operating Systems
Hard Reservation
28 Real-Time Operating Systems
Firm Reservation
29 Real-Time Operating Systems
Soft Reservation
30 Real-Time Operating Systems
Commercialized Linux/RK: TimeSys Linux
Resource kernel and QoS Support TimeSys Linux/GPL
Basic TimeSys Linux kernel
guaranteed, timely and enforced access to
Full preemption at the kernel level,
CPU cycles and network bandwidth prioritized interrupt handlers, unlimited
SMP support with QoS Reservations priorities, ...
TimeSys Linux/Real-time
Fully preemptive kernel Support priority inheritance and a POSIX-
Fixed-priority scheduling (POSIX-compliant) based high-resolution timer API
High-resolution timer and clock support TimeSys Linux/CPU
Support CPU reservation, which gives a
(microsecond resolution) thread, process, or process group
Periodic processes exclusive use of the CPU.
TimeSys Linux/Net
Message queues
Support network bandwidth reservation
Priority inheritance and priority ceiling protocol to guarantee that your thread or process
will get the bandwidth it requires,
emulation support to avoid unbounded priority regardless of network activity in other
inversion processes
TimeSys Linux GPL: Downloadable from
sourceforge.net/projects/timesysgpl
31 Real-Time Operating Systems
Sources
Victor Yodaiken, The RTLinux Manifesto, In Proceedings of the
5th Linux Expo, 1999
Victor Yodaiken, Cort Dougan, Michael Barabanov,
RTLinux/RTCore dual kernel real-time operating system, FSM
Labs White Paper, 2004 [ available from
http://vyodaiken.com/papers-and-talks/ ]
QNX Neutrino RTOS System Architecture [ available from
qnx.com/download/ ]
http://www.lynx.com/products/real-time-operating-
systems/lynxos-rtos/
http://www.VxWorks.com/collateral
Raj Rajkumar, Kanaka Juvva, Anastasio Molano and Shui Oikawa,
Resource Kernels: A Resource-Centric Approach to Real-Time
Systems, In Proceedings of the SPIE/ACM Conference on
Multimedia Computing and Networking, January 1998.
32 Real-Time Operating Systems
