Performance Evaluation of Reservation MAC Protocols under ... - ISPLC

18 downloads 3991 Views 245KB Size Report
reservation MAC protocols for PLC networks to deal ... Communications networks using reservation MAC ... Reservation domain, specifying a data unit or a.
Performance Evaluation of Reservation MAC Protocols under Selective Disturbances in Broadband PLC Networks Halid Hrasnica, Ralf Lehnert Chair for Telecommunications, Dresden University of Technology D-01062 Dresden, Germany E-mail: {hrasnica | lehnert}@ifn.et.tu-dresden.de Phone: +49 351 463-{33474 | 33945}, Fax: +49 351 463-37163 Abstract Broadband PLC access networks have to operate with a limited signal power, which makes them more sensitive to disturbances. We can also expect that the disturbances differently affect particular network segments and with it, they can have different influence on network stations, depending on their position in a PLC access network. We show that a such selective disturbance scenario significantly decreases network performance. Therefore, we propose a protection mechanism to be applied within reservation MAC protocols for PLC networks to deal with this problem. It can be concluded that the protected protocol variants completely eliminates negative influence of selective disturbances. 1.

can be achieved by using efficient methods for the network capacity sharing - Medium Access Control (MAC) protocols.

Introduction

Broadband PLC systems applied to the telecommunications access networks use a frequency spectrum up to 30MHz. However, the low-voltage supply networks are not designed for communications and they act as antennas, producing electromagnetic radiation and causing disturbances to other telecommunications services operating in this frequency range. Therefore, PLC networks have to operate with a limited signal power, which decreases data rates and makes PLC systems more sensitive to disturbances from the electrical power supply network, as well as from the network environment [1]. Well-known error handling mechanisms, such as Forward Error Correction (FEC), interleaving, and Automatic Repeat request (ARQ) are applied to PLC systems to solve the problem of transmission errors [2]. However, application of these mechanisms consumes a part of the transmission capacity (due to coding overhead and data retransmissions) and further decreases the data rates. On the other hand, PLC networks have to provide a very good utilization of its shared transmission medium (Figure 1) keeping also a sufficient Quality of Service (QoS) for different telecommunications services. This is important because the broadband PLC access networks have to compete with other access technologies (e.g. DSL, CATV). Both good network utilization and sufficient QoS

Figure 1: PLC Access Network Communications networks using reservation MAC protocols are suitable to carry hybrid traffic (mix of traffic types caused by various telecommunications services) with variable data rates [3]. The reservation protocols allow also realization of various QoS guarantees and ensure a near to full network utilization. On the other hand, PLC networks are expected to be disturbed by different noise sources [1], [2]. In the case of reservation protocols, the transmission is controlled by a central unit (base station), which is also favorable for realization of an efficient fault management. For all these reasons, we propose application of reservation MAC protocols in PLC access networks. A reservation of the transmission capacity is done for a particular user or a service. For this purpose, one or more transmission channels are reserved for the signaling transmission of user requests to the base station and acknowledgements (transmission rights) from the base stations. After the reservation procedure is finished, the data transmission is contention-free. However, reservation signaling procedure can also be affected by the disturbances, which causes repetition of the signaling procedure and longer transmission delays. Particularly, the

disturbances can influence a PLC network selectively (e.g. only a part of the network), causing unwanted packet collisions, even if the reservation protocols are applied. The aim of this investigation is performance analysis of the reservation MAC protocols under selective disturbances and consideration of possible solutions to protect the reservation protocols against a selective disturbance scenario. The paper is organized as follows: First we describe components of the reservation MAC protocols (Sec. 2) and influence of disturbances on the protocol performance (Sec. 3). There, we propose a mechanism to improve robustness of the reservation protocols against selective disturbances. In Sec. 4, we describe simulation model used for performance evaluation and present achieved results, presenting that selective disturbances have a big negative influence on the network performance. However, we showed the negative influence of selective disturbances can be successfully avoided by proposed protection mechanism to be implemented within reservation MAC protocols. 2.

Reservation MAC Protocols

2.1. Components of reservation MAC protocols A reservation MAC protocol merges several functions which are necessary for realization of the medium access, including the entire signaling procedure between multiple network stations and a base station (Figure 1). The following four protocol components can be defined: Reservation domain, specifying a data unit or a time period for which the reservation is carried out, Signaling procedure, describing an order of events for the exchange of signaling messages between the network stations and the base station, Access control, ensuring collision-free medium access for multiple stations, and Signaling MAC protocol, applied in the part of the network capacity allocated for realization of the signaling procedure (e.g. signaling channel). A signaling MAC protocols controls transmission of the requests in the uplink direction (from network stations to a base station), as well as transmission of acknowledgements and transmission rights in the downlink (from the base station to network stations). Disturbances can affect the signaling channel causing a interruption of the signaling procedure and its repetition. The selective disturbances do not have a particular impact on the signaling protocols and its influence is same as in the case of non-selective disturbances. So, if the base station has not received a transmission requests it does not answers and the affected network station will repeat the request, independently on the type of disturbance which corrupted the requests. If the

acknowledgement from the base station is corrupted, the network station will repeat the request, as well. Thus, in case of both selective and non-selective disturbances there is no influence on the collision-free data transmission, ensured by a reservation MAC protocol. In this investigation, we consider a network applying so-called extended hybrid two-step signaling protocol, described in [4]. The extended protocol includes piggybacking, signaling over data channels, and dynamic backoff mechanism for stabilization of the random access [5] (for hybrid part of the protocol and access to the signaling channels for signaling purposes). 2.2.

Signaling procedure and access control

Acknowledgements and allocation messages can be transmitted separately (e.g. in CPRMA protocol, [3]) or in the same packet. In the first case, the station receives an acknowledgement immediately after the base station received its transmission request (Figure 2). The acknowledgement informs the requesting stations only that its transmission request has been arrived the base station. The allocation message, containing information about access rights, is transmitted later, directly before the transmission starts. In the second case, both acknowledgement and allocation messages are transmitted jointly, immediately after the transmission request is received by the base station. The transmission of only one control message per request is the more efficient solution because of the following reasons: downlink of the signaling channel(s) is less loaded and error probability for the control messages decreases because there is a less number of transmitted control messages. Therefore, we consider the signaling procedure with joint control messages in this investigation.

Figure 2: Reservation Signaling Procedure As is mentioned in Sec. 1, broadband PLC access networks are expected to ensure realization of various telecommunications services. For this purpose, network resources can be allocated for particular services carrying their data packets. So, in the logical channel structure, presented in Figure 3, a transmission channel can be allocated, usually in a dynamical manner, for transmission of various service classes. Additionally, for realization of the reservation procedure it is necessary to allocate a

certain portion of the network transmission resources (signaling channel, Sig). A transmission channel can also be temporarily free (Idle), or it can be reserved (Res.) for a special purpose. The logical transmission channel represent the accessible portions of the network resources, provided by a multiple access scheme (e.g. TDMA, FDMA, etc.).

Figure 3: Channel State Diagram A data channel can be allocated to be circuit or packet switched (CS and PS), which depends on the traffic characteristics of service classes using the transmission channel. So, if we consider a classical telephony service, for this service class it is suitable to allocate the circuit switched channels, which remain allocated for a voice connections for its entire duration. The CS channels can also be allocated for various data connections in accordance with the per-burst reservation domain [6]. In this case, the allocated channels are not released after the end of a connection, but they remain allocated until end of a data burst. However, this is not an efficient reservation method because of the transmission gaps, but it can be necessary to ensure required QoS guarantees for specific services. On the other hand, packet switched channels can be allocated only for transmission of one data packet, leading to a per-packet reservation domain [7]. After the transmission is completed, the channel is free and can be used for a new transmission, either as a packet or a circuit switched channel. 2.3.

However, the traffic conditions in the network can change; either because of arrival of connections with higher priorities than currently admitted connections in the network, or because of the variation of available data rate in the network caused by disturbances. In both cases, it can happen that connections with lower priorities have to be postponed to ensure an immediate transmission of data from connections with higher priorities. Then, an additional allocation message informing the network stations about the rescheduling of their connections has to be sent by the base station. The additional allocation message can also be corrupted by the disturbances which can affect a PLC network selectively (Sec. 3), causing unwanted transmission collisions in the network. To avoid this situation, we implement a distributed access control mechanism which ensures that the waiting stations, i.e. stations which have already received an acknowledgement from the base station (in accordance with the signaling pürocedure with joint control messages), observe the situation in the network and accordingly calculate by themselves a new time for the beginning of a transmission (Figure 4). The acknowledgement contains a number of data slots which has to be passed by the station before it starts to send (SP). For each slot to be passed that is used by other network stations, which made the reservation earlier, the counter is decremented by one. If the counter is zero the station can start the transmission in the next available data slot.

Access to the packet-switched channels

In the case of a service using circuit switched channels, the reservation procedure seems to be simple. After a CS request, the allocation message sent by the base station contains the identification number of one or more transmission channels which are allocated to a particular station for the entire duration of the connection. After the connection is completed, the used channels are again free. However, in the case of packet switched channels, there could be a time period between the reception of a request from a network station and the beginning of the actual data transmission (Figure 2). This can happen because some data from other network stations, which already complete the signaling procedure, have to be transmitted at first and these transmissions are not yet finished.

Figure 4: Distributed Access Control Mechanism Thanks to the distributed access control mechanism, a waiting station always keeps information about the number of data segments which have to be transmitted by other stations before it starts to send, independently of the changing network data rate This also ensures that a waiting

station starts the transmission immediately after a previous transmission is completed which improves the network utilization, as well.

50% stations with the probability 0.15, and only 25% stations in the network are affected by an impulse with the probability 0.10.

3.

Influence of Disturbances

3.3.

3.1.

Disturbance model

Selective disturbances in a PLC network, acting only in particular network segments, can desynchronize the distributed access control (Sec. 2) causing unwanted transmission collisions. To deal with the problem of the selective disturbances and their negative effects, the distributed access control mechanism can be protected by provision of an additional information about data-slot order, which prevents the unwanted transmission collisions. We implemented this mechanism in the following way: In every time-slot there is a sequence number determining actual order of available data slots. Thus, a network station which was not able to recognize one or more data slots, because it was affected by a disturbance, can update the slot counter (SP, Figure 4) and is again ready to access the medium without causing the collisions. In the case that a network station is not able to recognize data slots within a time-slot when it should start the transmission, it does not start to transmit and applies the reservation procedure from the beginning. In this way, the transmission is postponed, causing also additional transmission delay, but an unwanted collision is avoided, as well.

Transmission errors (disturbances) in PLC networks are mainly caused by asynchronous impulsive noise [1]. The disturbances caused by this type of noise can be represented by a so-called on-off disturbance model with the following two states: OFF - the channel is disturbed, no transmission is possible, and ON - the channel is available. These two states can be modeled by two random variables which represent interarrival times of the impulses/disturbances, moving the channel into the state OFF, and the disturbance duration. Both random variables are negative exponentially distributed in accordance with behavior of the impulsive noise [8], [9]. In this investigation the mean interarrival time of the impulses/disturbances is 200 ms and the mean duration of a disturbance impulse is 100 µs. We assume that the noise impulses with duration shorter than 300 µs do not cause errors because they can be handled by physical network layer or applied correction codes [2]. 3.2.

Selective disturbances

In our previous investigations (e.g. [1], [2], [4], [5]) we considered PLC networks where a disturbance affects all stations in a network. However, the length of typical PLC access networks is up to several hundreds meters. Thus, we can expect that the disturbances can differently affect particular network segments (Table 1); e.g. depending on the position of noise source, protection of powerline grids in different network sections, etc. Table 1: Parameter of Selective Disturbance Model Part of affected stations / %

Probability

100

0.50

75

0.25

50

0.15

25

0.10

In accordance with the disturbance model, chosen to represent selective disturbances in this investigation, all network stations are affected by a noise impulse with the probability of 0.5 (Table 1). 75% of the network stations are affected by a disturbance with the probability 0.25,

4.

Robustness against selective disturbances

Performance Evaluation

4.1. Simulation Model To investigate the protocol performance, we use a generic simulation model, described in [1]. The simulation model implements an OFDMA/TDMA (Orthogonal Frequency Division Multiple Access/Time Division Multiple Access) scheme, presented in [5]. There are 15 bidirectional transmission channels (64 kbps each) provided by the OFDMA in the network model and one of them is reserved for signaling. The TDMA component is applied to each transmission channel dividing them into time-slots, with a duration of 4 ms. Each time-slot can carry a 32 bytes data segment; 4 bytes are reserved for header information and 28 for payload. Accordingly, net data rate of a transmission channel amounts to 56 kbps. The user packets (IP packets) to be transmitted are reassembled into the data segments. The disturbances are modeled independently for each transmission channel in accordance with the on-off and selective disturbance models, presented in Sec. 3. We analyze internet traffic as the mainly used service in the PLC access networks. The mean packet size is set to 1500

bytes according to the maximum size of an Ethernet packet. The mean interarrival time of packets represents user requests for download of WWW (World Wide Web) pages and it is chosen to be 4,8 s. So, the average data rate per subscriber amounts to a relatively low value of 2,5 kbps. However, in other studies considering internet based data transmission in the uplink direction, offered network load per user is even lower (662,5 bps [10]). To increase network load, we increase number of network stations from 50 to 500. Both random variables, packet size and interarrival time, are modelled as negative exponentially distributed random variables. 4.2.

Network utilization and data throughput

Network utilization (Figure 5) is calculated as the relation between used network capacity for the data transmission and common network capacity. Note, that only successfully transmitted data packets are calculated for used network capacity and disturbed packets, which are retransmitted, are considered as lost network capacity. In the error-free network, a near to full network utilization of approximately 93% is achieved. The remaining 7% of the network capacity are allocated for the signaling (one of 15 channels) and it is never used for the data transmission. In the network model applying non-selective on-off disturbance model (Sec. 3), maximum network utilization is decreased to approximately 83%.

network utilization decreases to approximately 83% (Figure 5). Thus, we can conclude that the proposed protection mechanism operates successfully achieving almost the same performance, as it is the case in the network with only non-selective disturbances. In the case of protected protocol, there is an additional control message form the base station informing the network stations about the actual sequence number of the data slots (Sec. 3). If a station missed the synchronization and does not receive the sequence number message it is not able to transmit within the time-slot. Thus, the signaling procedure with joint control messages (Figure 2) is extended for an additional control message, used in the case that a station lost the synchronization, which practically leads to a signaling procedure with more than one control message. Therefore, error probability that a control message is disturbed increases and the network utilization decreases slightly (Figure 5). So, we can conclude that application of the signaling procedure with separated control messages (Figure 2) does not significantly decreases the network performance. However, in a network with worse disturbance conditions it is expected that the network performance is more decreased. Data throughput is calculated as the relation between the realized and offered data rates for each network station. It can be concluded that the data throughput decreases with the increasing number of network stations (increasing network load) and follows the results achieved for the network utilization (Figure 6). So, below network saturation point for different disturbance scenarios, the data throughput is 1, whereas beyond the saturation point it decreases. As expected, the highest data throughput is achieved in the error-free network, followed by network with non-selective disturbances and network with selective disturbances with implemented protection mechanism. Significantly worse performance is achieved in the case of non-protected protocols in the network under selective disturbances.

Figure 5: Network Utilization - Selective Disturbances If the network under selective disturbances (Table 1) is considered we can observe a dramatically decrease of the network utilization. In this case, the distributed access control mechanism (Sec. 2, Figure 4) is desynchronized, as described in Sec. 3, causing unwanted packet collisions. Accordingly, a collision-free data transmission is not ensured any more and network performance remembers on typical behavior of contention MAC protocols [1], [5], [11]. On the other hand, if we apply the protection mechanism against selective disturbances, described in Sec. 3, the

Figure 6: Average Data Throughput

4.3.

Transmission delay

Figure 7 presents mean transmission delay of user packets, calculated as a time from the packet generation and the succeeding transmission request until successful packet transmission including all possible retransmissions of the requests and corrupted data packets. As expected, the shortest transmission delay is achieved in the error-free network. A big difference can be observed if we consider the network under selective disturbances, where the transmission delay is significantly longer. Application of the protected protocol improve clearly the transmission delay, achieving the same result, as in the network with only non-selective disturbances.

References [1]

[2]

[3]

[4]

[5]

Figure 7: Mean Transmission Delay 5.

[6]

Conclusions

We investigated performance of a broadband PLC access network under influence of selective disturbances, differently affecting particular network segments. We considered a PLC network using a per-packet reservation MAC protocol, which applies a simple signaling procedure with a distributed access control mechanism, ensuring low error probability in the network and always a collision-free data transmission. Network performance observed under influence of the selective disturbances decreases dramatically because of desynchronization of network stations, which increases number of unwanted packet collisions. To solve this problem, we implemented a protection mechanism within reservation MAC protocol which successfully avoids the negative influence of the selective disturbances. The protection mechanism provides exchange of additional control messages which inform network station about actual order of data slots and ensure their synchronization in every time-slot. The increase of number of the control messages increases the number of exchanged control messages in the network and error probability, as well. However, we can conclude that this does not decrease network performance significantly.

[7]

[8]

[9]

[10] [11]

H. Hrasnica, A. Haidine, Modeling MAC Layer for Powerline Communications Networks, Internet, Performance and Control of Network Systems, part of SPIE's symposium on Information Technologies, Boston MA, USA; 5-8 November, 2000 H. Hrasnica, R. Lehnert, Performance Analysis of Error Handling Methods Applied to a Broadband PLC Access Network, SPIE’s int. symposium ITCom2002: Showcasing Communication, Networking, Computing, and Storage Technologies and Applications – conference: “Internet, Performance and QoS” – Boston MA, USA; July 29th – August 1st, 2002 I. F. Akyildiz, J. McNair, L. C. Martorell, R. Puigjaner, Y. Yesha, Medium Access Control Protocols for Multimedia Traffic in Wireless Networks, IEEE Network - July/August 1999 H. Hrasnica, R. Lehnert, Performance Analysis of Two-step Reservation MAC Protocols for Broadband PLC Access Networks, 6th International Symposium on Power-Line Communications and its Applications (ISPLC2002), Athens, Greece; 27-29 March, 2002 H. Hrasnica, R. Lehnert, Extended ALOHA and Hybrid Polling Reservation MAC Protocols for Broadband Powerline Communications Networks, XVIII World Telecommunications Congress (WTC2002), Paris, France; 22-27 September, 2002 R. Kalden, I. Meirick, M. Meyer, Wireless Internet Access Based on GPRS, IEEE Personal Communications; April 2000 R. Fantacci, F. Innocenti, Performance Evaluation of a Modified PRMA Protocol for Joint Voice and Data Packet Wireless Networks, IEEE Transactions on Communications, Vol. 47, No. 12, December 1999 M. Zimmerman. K. Dostert, The Low Voltage Power Distribution Network as Last Mile Access Network – Signal Propagation and Noise Scenario in the HF-Rang, International Journal of Electronics and Communications (AEU) 54, No. 1, 2000 M. Zimmerman, K. Dostert, An Analysis of the Broadband Noise Scenario in Powerline Networks, International Symposium on Powerline Communications and its Applications (ISPLC2000), Limerick, Ireland; April, 2000 C. Trabelsi, H. K. Choi, Simulation and Modeling of MAC Protocols for Wireless Packet Data Networks Using Web Traffic, 2nd IS-95 UMTS Workshop R. Rom, M. Sidi, Multiple Access Protocols Peformance and Analysis, Springer-Verlag New York Inc.; 1990