Communication Model

In the communication model we will specify three things

• The communication topology

• The communication service

• The failure model of the service

Before this can be done it is essential to determine on what layer of the architecture the model is situated because different layers provide different services, exhibit different fail-ure modes, and have different topologies.

4.2.2.1 Layer of the model

As we are going to present those aspects of the PCF that deal with maintaining and proc-essing the polling list as part of the protocol stack, our model describes a communication service within the data link layer. The interface of this service is situated above the DCF (cf. Figure 4-10). Beneath this interface, framing, addressing, and detection of transmission errors are provided as services of the underlying sub-layers, while the polling functionality, recovery from transmission errors, and other services are assumed to be located above this interface. Also the inter frame timing is assumed to be handled by the underlying DCF layer, which allows us to focus our model of the PCF to those aspects not addressed in the standard.

Physical Layer

Framing / Addressing / Error Detection Distributed Coordination Function

Modeled Interface Polling / Re-Transmission / other Services

Figure 4-10. Layer of the model

Actually, we are forced to build the physical implementation above a DCF MAC layer encapsulated in a network interface card (NIC) that does not allow using the special inter frame spaces for PCF frames. All frames transferred to the NIC for transmission, will be

sent out with the complete DCF access procedure (including random backoffs) and frames with individual addresses will incur MAC level re-transmissions. This is because the MAC functions are time critical and are implemented in the firmware on the NICs. NIC vendors keep the source code for their firmware under disclosure, so enhancements cannot be added on that layer, nor can the interfaces be extended. However, this is not too much of a problem for the implementation. The timing of the protocols presented in the following only bases on the assumption that frames are not arbitrarily delayed before being transmit-ted on the MAC and that the probability of message losses does not grow arbitrary due to collisions. Both assumptions hold even if the PCF is implemented on top of the full DCF access procedure. The PCF still ensures that stations will not observe a busy medium when they are allowed to transmit a frame and that no collisions occur because of simultaneous access.

4.2.2.2 Topology

We model the wireless broadcast medium as a set of virtual point-to-point links that may have different frame loss rates and observe independent frame losses. This means that a broadcast frame send simultaneously over all the virtual links may be lost on some links and received on others. It does not mean a statistical independency of the frame losses.

Assuming virtual point-to-point links is a more realistic model than assuming a homoge-nous broadcast medium where all stations observe the same link quality and frames losses are perceived consistently. In wireless communication the probability of frame losses de-pends on the relative spatial location of sender and receiver and on the kinds of objects in their surrounding. Hence it is a property that may be different for any pair of stations.

We are considering communication within a single BSS in which the AP controls medium access through polling. As explained above, in such a structure all communication takes place between the AP and the clients; clients do not communicate directly among them-selves. Thus, only the virtual links between the AP and the clients are relevant in our model, which leads to a virtual star topology being imposed on the BSS (cf. Figure 4-11).

To some extend, this virtual topology is less dynamic than the real, complete topology of the BSS; yet, links can still go down and have varying loss rate as will be discussed later in the section. The set of virtual links makes up the communication sub-system, which pro-vides the communication service to the protocol entities.

AP STA2

STA1

STA4 STA3

virtual point-to-point link

Figure 4-11. Virtual star topology of the network

4.2.2.3 The Communication Service

In the following we define the communication service through a sequence of properties (Schemmer and Nett 2003b). For these definitions, we assume that all frames sent can be distinguished. Note that this is only assumed for sake of the definitions, but that neither the service nor its user is required to provide such a unique distinction. We say that a station si

sends frame m if the service user at si requests the communication service to transmit frame m, and we say that station si receives frame m if the communication service at si sends a receive indication for m to the service user. We denote by rec(m) the set of intended re-cipients of m.

Property 4-4 (Validity). There exists a constant δframe such that for all stations si, sj and all frames m, if si sends m at t and sj ∈ rec(m) and sj is correct throughout [t,t+ δframe], then sj receives m.

The Validity defines the very basic property of the service; that is, the transmission of frames. It states that when a station sends a frame, all intended recipients will receive it.

Property 4-5 (Integrity). For all stations si, all messages m, and all times t, if si receives m at t, there exists a time t' and a station sj such that t' < t and sj sent m at t' and si ∈ rec(m). Furthermore, for all stations si, all messages m, m', and time t, t', if si receives m and m' at t and t' respectively and t ≠ t', then m ≠ m'.

This property requires that the service does not modify frames, duplicate frames, or for any other reason indicates frames that were not sent. The following facts render this assump-tion plausible:

• The modeled sub-layer does not perform re-transmissions. As duplicate frames are usually caused by re-transmissions, this is not likely to happen here.

• Regarding modification of frames, the IEEE 802.11 Standard provides a CRC-32 (cyclic redundancy checksum) as a frame check sequence to ensure integrity. As-suming that within the modeled layer only non-code value failures occur, any modification can be detected and the service will not indicate the erroneous frames to its user. Thus, each modification that may occur during physical transmission is detected and results in no frame being delivered. If the integrity provided by the CRC-32 is not sufficient, further error detecting codes can be added, which how-ever is not in the scope of this thesis.

Property 4-6 (FIFO). For all stations si, sj and messages m, m': If si sends m before m' and sj receives m and m', then sj receives m before m'.

The FIFO property requires that frames from the same sender be delivered in the order they have been sent. On a single LAN, with only a single path from transmitter to receiver, distortion of the FIFO order only may occur if re-transmissions are used. As the modeled sub-layer does not perform re-transmissions, the FIFO order is always maintained.

Property 4-7 (Timeliness). There exists a known constant δframe such that for all stations si, sj, messages m, and times t, t': If si sends m at t and sj receives m at t', t' - t ≤ δframe

While Validity already requires that correct stations receive frames in time, timeliness en-sures that no station, even a faulty one, receives a frame late. As explained above arbitrary delays are unlikely on the sub-layer on which our model resides for the following reasons

• There are no retransmissions, so only the delay of a single physical transmission is considered

• The medium access delay is bounded since under the PCF stations need not wait for the medium to become idle.

Obviously, timeliness is only ensured as long as frames are sent at a limited rate. In par-ticular, if a second frame is sent while the transmission of the first frame is still in pro-gress, a queuing delay at the local station will be the consequence. We therefore assume that the communication service sends a status indication to its user whenever the transmis-sion of a frame ceases. So, the user is able to synchronize with the rate of the communica-tion service. Figure 4-12 illustrates this idea: when the transmission of m ends, the com-munication service sends stat_ind. Reacting to this signal and sending the next frame takes at most δsched time units. The figure also makes clear that the indication is not expected to be sent at the same time when sj receives frame m; for example, the propagation delay has to be considered, which we assume to be small however.

Figure 4-12. Timing of the frame transmission

The properties presented in this clause define the service provided by a correct communi-cation link; in the following clause, we will define how a faulty link may deviate from this service.

4.2.2.4 Failure Model

For the communication service we assume omission/crash failure semantics. This means that if a station sends a frame, not all of its intended recipients may receive it. Thus, with omission failures, the validity property is not always fulfilled by the actual service of the link. We assume that omission failures may be perceived inconsistently; that is, if a frame has more than one intended recipient, some of them may receive the frame and some may not. This corresponds to the idea of modeling the broadcast medium as a set of virtual point-to-point links where a frame sent simultaneously over several links may be lost on some links and be received on the others. A crash failure of a link means after certain point of time on no frame will be received on that link. Link crashes may be due to the mobility

of the stations; in particular, if a station moves thus far from the AP that no more frames will be received. Inconsistency of the failures raises another important issue, especially in cooperative applications. Inconsistent failures leave the intended recipients of the frame in inconsistent states, which may lead to inconsistent actions of the stations.

Summarizing we assume a synchronous system model with omission failures. This holds for both the process and the communication model. For both, a timely predictable behavior is assumed since the services have timing specification and do not exhibit timing failures.

While our CPU scheduling achieves this for the process model, the basic medium proper-ties under the PCF ensure it for the communication model. We do not assume a limit on the number of omission failures as part of the basic system model. Thus, the general model presented so far does not allow realizing a reliable and timely communication between the stations. In fact, in a wireless network with mobile stations, this cannot be ensured for all stations and at all times. For which stations such a service can be provided and for which not depends on the dynamically changing link properties, which will be considered in the following sub-section.