Real-Time Systems. Design Principles for Distributed Embedded Applications. Herman Kopetz. Second Edition (811374), страница 55
Текст из файла (страница 55)
Since we cannot predict the instant of message delivery, rateconstrained communication systems are not deterministic according to the definition of determinism given in Sect. 5.6.1.7.4.1Token ProtocolOne of the early rate-constrained protocols is the token protocol controlling theaccess to a multi-access local area networks (LAN). In a token system, the right totransmit is contained in a special control message, the token. Whoever has the tokenis allowed to transmit.
Two time parameters determine the response time of a tokensystem, the token-hold time THT, denoting the longest time a node may hold thetoken, and the token-rotation time TRT, denoting the longest time for a full rotationof the token. The maximum TRT is the product of the number of nodes and theTHT of each node, which determines the guaranteed bandwidth allocated to a node.A serious failure in any token system is the loss of the token, e.g., if the station thatpossesses the token fails.
In such a situation, the network traffic is disrupted untilsome other node detects the silence by monitoring a time-out, and generates a newtoken. A token system can be structured as a bus or a ring. Token rings werestandardized by IEEE standard 802.5. They have been in wide use some years ago.7.4.2Mini-slotting Protocol ARINC 629In the mini-slotting protocol ARINC 629 [ARI91], the access to a shared bus iscontrolled by two time-out parameters, the synchronization gap SG controllingthe entrance to a distributed waiting room, and the terminal gap TG controlling theaccess from the waiting room to the bus. The synchronization gap SG is identicalfor all nodes, whereas the terminal gap TG, the personality timer, is different foreach node and must be a multiple of the propagation delay (called a mini-slot).
Thefollowing relation holds between these time-outs: SG > Max{TGi} for all nodes i.ARINC 629 is thus a waiting-room protocol similar to the bakery algorithm ofLamport [Lam74]. In the first phase, the set of nodes that wants to transmit amessage is admitted to the distributed waiting room if there is silence on the bus fora longer duration than the synchronization gap SG. A node that has entered the1827 Real-Time Communicationwaiting room and senses silence on the bus for a duration that surpasses its personalterminal gap TG starts transmitting its message. This protocol logic guarantees thata node cannot monopolize the bus, since even the node with the shortest terminalgap TG (the highest priority node) is allowed to send a second message only afterall other nodes that are in the waiting room have finished their message transmission. Typical values for the time-out parameters on a 2 Mbit/s channel are: terminalgap (determined by the propagation delay): 4–128 ms, synchronization gap SGlonger than the longest terminal gap.
The ARINC 629 protocol is used on theBoeing 777 airplane.7.4.3Avionics Full Duplex Switched EthernetAvionics Full Duplex Switched Ethernet (AFDX) is a rate-constrained protocolbased on switched Ethernet. Whereas the message format and the physical layer ofAFDX is in agreement with the Ethernet standard IEEE 802.3, the protocolallocates a statically defined bandwidth to each sender on a virtual link basis.A virtual link connects a sender with a specified number of receivers. An AFDXswitch guarantees that1.
The delivery order of messages traveling on a virtual link is the same as the sendorder.2. A minimal bandwidth and a maximum transmission latency and jitter is guaranteed on a virtual link basis.3. There is no data loss due to buffer over-subscription in the switch.The configuration table of the switch contains state information for every virtuallink and enables the switch to protect the network from nodes that try to overloadthe network. The system integrator establishes the virtual links and sets the connection parameters.
AFDX has been standardized under ARINC 664 and used in theAirbus A 380 and in the Boeing Dreamliner B787.Table 7.2 shows typical values of the virtual link latency distribution, i.e., thejitter, on an AFDX configuration for the A 380 [Mil04]. This jitter will cause asignificant action delay (see Sect. 5.5.2 and Problem 5.16).7.4.4Audio Video BusPhysical connections in multimedia systems are predominantly unidirectional andpoint-to-point, resulting in substantial wiring harnesses. In order to simplify thewiring harnesses, special multi-media protocols have been developed.
Some ofthese special protocols are incompatible with standard IT protocols such as Ethernet. On the other side, standard switched Ethernet does not provide the temporalquality of service needed in multi-media applications.7.5 Time-Triggered Communication183A communication system for audio-video streaming must support the followingtemporal quality of service requirements:1. It must be possible to precisely synchronize multiple audio video streams thatare generated at different physical locations, e.g., for lip synchronization ormixing content.
The required precision of synchronization is in the microsecondrange.2. The worst-case transport delay of a multimedia data stream, including bufferingdelays at source and destination, must be bounded. The duration for the dynamicswitching from one video stream to another video stream must be in themillisecond range.3.
The communication resources that are dynamically allocated to a multi-mediastream must remain available for the duration of a session. This requires adynamic resource reservation schema.The IEEE 802.1 audio/video bridging (AVB) task force develops a set of protocolsbased on the Ethernet standard that meet the aforementioned requirements.7.5Time-Triggered CommunicationIn a time-triggered communication system, the sender and receiver(s) agree a priorion a cyclic time-controlled conflict-free communication schedule for the sending oftime-triggered messages. This cyclic communication schedule can be expressed inthe cyclic model of time (see Fig.
3.9 in Sect. 3.3.4), where the send and receiveinstants of a message, the message cycles, are represented by a period and phase. Inevery period, a message is sent at exactly the same phase. Since the communicationsystem can be aware of the schedule, it can allocate the resources such that a timetriggered message is transported without any intermediate delay or buffering.In some sense, time-triggered communication resembles time-controlled circuitswitching (TCCS), where a time-controlled dedicated channel between a sender andreceiver is established for the short duration of a single message transport.Example: A coordinated set of traffic lights along a road that establishes periodically agreen wave is a good metaphor for time-controlled circuit switchingWe can distinguish the following three types of time-controlled circuit switching:1. Collision Avoidance Time-Controlled Circuit Switching (CA-TCCS).
In CATCCS, it is assumed that there are two message classes, the scheduled timetriggered messages and the sporadic event-triggered messages. The switch,knowing beforehand the conflict-free schedule of the time-triggered messagescan shift the event-triggered messages such that conflicts between the eventtriggered and time-triggered messages are avoided.2. Preemptive Time-Controlled Circuit Switching (P-TCCS). In P-TCCS, it isassumed that there are two message classes, the scheduled time-triggered1847 Real-Time Communicationmessages and the sporadic event-triggered messages.
In case of a conflictbetween an event-triggered message and a time-triggered message, the switchpreempts the event-triggered message and sends the time-triggered messagewith a small delay and minimal jitter.3. Collision Tolerant Time-Controlled Circuit Switching (CT-TCCS). In CTTCCS, it is assumed that there are two message classes, the scheduled timetriggered messages and other uncontrollable messages or interfering signals, asituation characteristic for wire-less scenarios. The time-triggered communication controller will send multiple replicas of the same time-triggered message ondiverse frequency channels at different preplanned instants, hoping that one ofthe replicas will reach the receiver unharmed.Within the given fault hypothesis, time-triggered communication is deterministicaccording to the definition of determinism in Sect. 5.6.1.
The sparse model oftime ensures that messages that are sent at the same active interval of the sparsetime base over independent (replicated) channels will arrive at the receivers at thesame future active interval of the sparse time base. The jitter is bounded by theprecision of the clock synchronization, which is usually in the sub-microsecond range.Time-triggered control requires that the temporal control signals within adomain be derived from a single time source.
This time source can be the synchronized global time (this is preferable) or the period of a single leading process, i.e.,a process that establishes a basic period autonomously. In case a leading processestablishes the time, the temporal control signals of all other cycles must be derivedfrom the period of this leading process. The generation of the schedules is simple, ifall periods are in a harmonic relationship.Example: If, within single synchronization domain, two processes are activated byunrelated control signals from two different unsynchronized time sources, then the phaserelationship between these two processes will vary such that eventually the process executions will overlap.The precise phase control of time-triggered communication makes it possible toalign the processing actions and communication actions tightly within a distributedtransaction and thus minimize the duration (the dead-time) of a control loop(see Fig.
![](https://s.studizba.com/z.php?f=/templates/si/i/noavatar.png&w=100&h=100&t=1"){kind=avatar}
