Model of Real Time Communication

Model of Real-Time Communication

The figure below shows the well-known model of distributed systems. The hosts are connected by a communication network or several interconnected networks. The top layers are simplified and all the entities above all the transport layer applications are called.

Architectural Overview

The source and destinations of every message are application tasks residing on different hosts. The network interface of each host contains an input queue and an output queue which can also be referred to as input/output buffers or simply buffers. For the sake of correctness, it is assumed that these queues are jointly maintained by two local servers: the transport (TP) handler and Network Access Control (NAC) handler. The TP handler interfaces with local applications and provides them with message transport services. The NAC handler interfaces with the network below and provides network access and messages transmission services to the TP handler.

Figure shows the data paths indicated by heavy arrows traversed by messages in and out of two hosts. The circles marked TPH and NACH are TP and NAC handlers, respectively. When requested to send a message by a local application task, the source TP handler places the message in that output queue. From there, each outgoing message is delivered to the network under the control of the source NAC handler. After the message has traversed to the network, the destination NAC handler places the message in the input queue and notices the destination TP handler. Then, the destination TP handler moves the message to the address space of the destination application task and notifies the application of the arrival of the message. (Liu, 2003, pp. 235-436)

Packets

Messages are fragmented into segments before transmission through communication network. Each segment is handled by the network as a basic transmission unit called a frame, a packet or a cell. The transmission of the unit is non preemptable. Example, a packet is a 53-byte cell in an ATM network. Queuing and propagation delay is also measured in terms of this time unit: the length of delay is measured in terms of the number of packets that can be transmitted in that length of time. Furthermore, the length of each transmission link is measured by the number of packets that can be transmitted within the length of time they take to reverse the link.

Real Time Traffic Models

In real time communication, the term ‘real time traffic’ means isochronous of synchronous traffic consisting of message streams which are generated by their sources on a continuing basis and delivered to their respective destinations on a continuing basis. Such traffic includes periodic and sporadic messages that require some guarantee for on time delivery. There are also aperiodic messages.

Each of these type of message is referred to as a message stream and is denoted by M_i for some index i to distinguish it from other messages. A message instance of packet arrives or departs at a point in time when the last bit in it arrives or departs.

Periodic and Aperiodic messages

The messages that are generated and consumed by periodic tasks are periodic messages. The characteristics of periodic messages are similar to the characteristic of their respective source tasks. The transmission of a periodic message is periodic task. For example, message streams carrying sensor data and actuator commands generated and consumed by digital controllers.

A periodic message is denoted by M_i by the tuple (p_i, e_i, D_i). It means that the inter-arrival times of instances in M_i are never less that the period pi of the message, the maximum length of instances in M_i is equal to e_i packets and each instance must be delivered to the destination within Di units of time from its arrival at the source. Di is the relative deadline to Mi. This traffic model is called the peak rate model in real time communication. (Liu, 2003)

There are also aperiodic message streams. The transmission of an aperiodic message is an aperiodic task. An aperiodic message stream does not have a relative deadline. However, it is efficient to keep the average delay suffered by an aperiodic message instances as small as possible.

Sporadic Messages

According to FeVe Model, a sporadic message M_i is characterized by a 5-tuple (p_i, p_i’, I_i, e_i, D_i). The parameters p_i, e_i, and D_i are the minimum inter-arrival time, maximum length and relative deadline of the instances in M_i respectively. P_i’ is the average inter-arrival time of the instances of M_i where the average is taken without waiting for the arrivals of length l_i. In a switched network, a switch transmits each packet without waiting for the arrivals of later packets in the same message instance. The FeVe model simplifies to M_i = (p_i, p_i’, I_i, D_i). the packets in Mi never arrive less than pi units of time apart, and their average inter-arrival time over any time interval of length I_i is p_i’. The maximum length of each instance of M_i is 1 and omitted in the tuple.

Performance Objectives and Constraints

We want to measure the performance of scheduling, synchronization, and flow control algorithms used for real time communication and the performance of the performance of the resultant communication system along two dimension: from the points of view of the user and the system. The user is concerned with the on time delivery of periodic and sporadic messages and the average response time of aperiodic messages.

Miss rate: the fraction of all message instances or packets that are delivered to their destinations too late.

Loss rate: gives the fraction of all message instances in the stream that are dropped en route for flow and congestion control reasons.

Invalid rate: combination or miss and loss rate (sum of miss and loss rate)

Delay Jitter: the variation in the delays suffered by different message instances or packets in the stream.

Buffer Requirement: a packet that arrives too early to be processed by the destination must be buffered. So, a larger delay jitter of a message stream means that more buffers must be provided by the stream.

Throughput: the rate of each message stream measures the throughput of the stream.

Real Time Connections and Service Discipline

According to the connection oriented approach, a logical simplex connection from the source to the destination is set up for the transmission of each message stream. All packets on each connection are sent along a fixed route.

Admission control and Connection Establishment

The use of a fixed route for each connection allows each switch en route to set aside the required bandwidth and buffer space for the connection so the network can provide some form of performance guarantee. This also enables the control over packet transmission to be done on a per connection basis.

The client declares the characteristics of the message stream and the required performance of the connection to request a connection. The characteristics are defined by parameters of message stream to be carried on the connection. These parameters are collectively called flow specification. The required performance is stated in terms of quality of service parameters like delay, jitter, and so on. The admission controller of each handler and switch along the chosen route uses these parameters as the basis of an acceptance test to determine whether to admit the connection. The connection is admitted if the requested quality of service is existing connections service by the handler or switch.

Packet Switched Networks

The figure ‘a’ illustrates a packet switched network. Figure ‘b’ shows m*m switch; it has m inputs and m output links, both called links 1, 2, . . . , m. The switch routes packets on its input links to its output links.

Each switching pattern can be represented by a permutation of m tuple (1, 2, . . . , m). for example, for a 4*4 switch, the 4 tuple (2, 4, 1, 3) means that a packet on input link 2 goes to the queue of output link 1, a packet on input link 4 goes to the queue of output link 2, and so on. The switch is nonblocking which means that every pattern represents a possible switching pattern. (Liu, 2003, p. 440)

Service Discipline

The combination of an acceptance test and admission control protocol, and synchronization protocol and a scheduling algorithm used for the purpose of rate control, jitter control and scheduling of packets transmission is called a service discipline. Rate control and jitter control serve the purpose of flow control for real time. Service discipline are of two types:

1. Rate allocating
2. Rate controlled

Rate allocating discipline allows packets on each connection to be transmitted at higher rates than the guaranteed rate provided the switch can still meet the guarantees to all other connections. A service discipline is rate controlled if it ensures each connection the guaranteed rate but never allows packets on any connection to be sent above the guaranteed rate. A bandwidth preserving server that claims all the background time is allocating.

References

Liu, J. W. (2003). Real Time System. Pearson Education, Inc. and Doring Kindersley Publishing Inc.