IOS Press
A priority-aware CSMA/CP MAC protocol for the all-optical IP-over-WDM
metropolitan area ring network
Jih-Hsin Ho
a, Wen-Shyang Hwang
b,∗and Ce-Kuen Shieh
aaDepartment of Electrical Engineering, National Cheng Kung University, Tainan 701, Taiwan, ROC E-mails: [email protected], [email protected]
bDepartment of Electrical Engineering, National Kaohsiung University of Applied Sciences, Kaohsiung 807, Taiwan, ROC
E-mail: [email protected]
Abstract. The paper proposes a priority-aware MAC (Medium Access Control) protocol for a core metropolitan area network in the next gen- eration Internet, which is an OPS (Optical Packet Switch) network that all-optically and directly transfers IP packets over a WDM (Wavelength Division Multiplexing) ring network. It uses the concepts of CSMA (Carrier Sense Multiple Access), CP (Carrier Preemption), and the priority mechanism to support all-optical and priority-aware transferring of the IP packets of the nodes in the WDM ring networks; the new MAC pro- tocol is named priority-aware CSMA/CP. Since the traditional IP provides the best effort service only, supporting IP packets with QoS transfer has become a crucial issue for multimedia transmission. Today, while the network bandwidth has grown dramatically, the kind of applications transferred are mostly high-bandwidth demanding multimedia transmissions. It is predictable that the end-to-end QoS will be an important area of study in the next generation Internet. This paper accordingly proposes an advanced mechanism for this, and gives a differential service model to analyze and simulate the average packet delay for each class.
Keywords: OPS, CSMA/CP, priority-aware, analysis and simulation
1. Introduction
Wavelength Division Multiplexing (WDM) [1,2], first developed during the late 1980s, provides tremendous bandwidth, up to OC-192 (10 Gb/s), and has total bandwidth on an optical fiber exceeding 20 Tbps. WDM-based solutions are therefore expected to appear in the next generation of access networks in metropolitan area networks.
However, harnessing this unprecedented bandwidth in the metropolitan network environment will require a WDM transmission protocol to efficiently transport IP packets across the data centric WDM-based MANs.
Due to the rapidly increasing services and user population on the Internet, IP packet traffic dominates the uti- lization of data networks. However, such packets are now transferred, switched, and manipulated through complex protocol stacks, such as IP/ATM/SONET/WDM, IP/HDLC/SONET/WDM, etc. Thus, the goal of merging and collapsing the middle layers of these stacks to reduce cost, complexity, and redundancy has become an important research issue [3]. In order to minimize the layering complexity and costs of SONET/SDH and ATM, the packet- based network traffic should be accommodated directly on the WDM network, which would to be an efficient and economical way to implement the next generation of the Internet. In this way, both the equipment cost and the management complexity related to electronic multi-layer solutions are significantly reduced in all-optical IP-over- WDM networks [4]. Additionally, since many WDM systems already have been deployed in Metropolitan Area
*Corresponding author. Tel: +886 7 3814526, ext. 5533; E-mail: [email protected].
0926-6801/07/$17.002007 – IOS Press and the authors. All rights reserved
Networks (MANs), the bottleneck of communications has been pushed from backbone networks to local access networks. As a result, applying WDM to Local Area Networks (LANs) has attracted much research interest.
Optical packet switching (OPS) [5] is one optical switching technique. Although synchronization of packets and hardware cost are the major drawbacks of OPS networks, this can be improved by transmitting high-speed optical base-band data with a sub-carrier header. The sub-carrier can be used to tag the high-speed base-band data payload with low-speed switching control information, so inexpensive and readily available low-speed electronics can be used at network nodes where high-speed data payload detection is unnecessary. In such a WDM network, one of the essential components will be the all-optical wavelength filter, such as a Semiconductor Optical Amplifier (SOA) filter [6]. The SOA filter is split into two output paths, one for control detection, where a low-pass-high-pass filter cascade passes only the sub-carrier; the other for base-band detection where a low-pass filter allows only the base-band data payload to pass.
The ring topology is chosen as an all-optical metropolitan backbone [7]. In addition to avoiding transmission collisions among nodes attempting to use the same network wavelength, an MAC protocol is necessary to arbitrate access to the wavelengths and detect or avoid collisions between nodes. Multi-token inter-arrival time (MTIT) access protocol was proposed in 1999 to support variable size IP packets over a WDM ring network whose archi- tecture is provided with fixed transmitters and fixed receivers [8]. To achieve all-optical communications, MTIT adopts a source removal policy for dropping packets from networks to prevent packet re-circulation. In 1999, Stan- ford University’s Optical Communication Research Laboratory (OCRL) [9–12] proposed the hybrid optoelectronic ring network (HORNET) using carrier sense multiple access with collision avoidance (CSMA/CA) protocol. HOR- NET utilizes optical-electronic (O/E) and electronic-optical (E/O) conversion to retransmit the bypassed packets back into the channel received, and employs a jamming signal mechanism to resolve the optical packet collisions.
Its data/packets regeneration can minimize the number of erbium-doped fiber amplifiers (EDFAs) and reduce am- plifier noise accumulation, however the O/E/O conversion will also constrain its transmission rate in the WDM backbone network. Marsan [13,14] proposed the Static Round-Robin (SRR), which is an almost optimal MAC protocol based on the time division multiplex (TDM) technique. SRR architecture has all-optical WDM multiple rings with tunable transmitters and fixed receivers. Due to the strict TDM design, the packet transmission to a node is constrained to using a particular fixed size slot. That scheme adopts the destination removal scheme to free the channel bandwidth to other nodes. In order to facilitate spatial reuse on the bandwidth of all optical ring networks, a carrier preemption access control protocol [15] has been first proposed by W.S. Hwang, etc. To avoid access collisions and use bandwidth more efficiently, W.S. Hwang, etc. proposed a novel MAC protocol that is based on the Carrier Sense Multiple Access and Carrier Preemption schemes named CSMA/CP [16]. An analytical model has also been developed to approximate the average transfer delay for the CSMA/CP MAC protocol [17,18].
Unfortunately, most of the research efforts in this area assume that connection is for a single class of traffic. In recent research, the WDM ring average delay analysis has been extended to multi-class services. In this paper, we will introduce service quality differentiation into the WDM ring network.
The first approach proposed to replace the best effort model is Integrated Services (Intserv) [19]. Using resource reservation and admission control (through protocols like RSVP), an application’s request for a certain level of performance can be guaranteed. But per-flow state information should be kept inside each router along the way in order to fulfill the service requirement of each flow. As a result, this approach encounters a scalability problem in its deployment. In order to overcome the scalability problem of Intserv, a relatively new approach: Differenti- ated Service (Diffserv) [20] has been proposed. Instead of providing end-to-end per-flow performance guarantees, Diffserv provides local (per hop) service differentiation for aggregated traffic with the same QoS requirement (per class). This goal is achieved by defining packets’ Per-Hop-Behavior (PHB) at each router. No state information about each flow is kept in the core of the network due to its per-class QoS model. In particular, two models of Diffserv have been identified: Absolute Service Differentiation and Relative Service Differentiation. The latter is receiving more attention because of its simplicity and its ability to be deployed incrementally [21].
This paper proposes a priority-aware MAC protocol of core metropolitan area network in the next generation Internet, which is an OPS network that all-optically and directly transfers IP packets over a WDM ring network.
It uses the concept of CSMA/CP MAC protocol [15,16] and the priority mechanism to support all-optical and
priority-aware transferring of the IP packets of the nodes in the WDM ring networks; the new MAC protocol is named priority-aware CSMA/CP. Since the traditional IP provides the best effort service only, the issue of supporting IP packets with QoS transfer has become a crucial issue for multimedia transmission. Today, while the network bandwidth has grown dramatically, the kind of applications transferred are mostly high-bandwidth demanding multimedia transmissions. It is predictable that the end-to-end QoS will be an important area of study in the next generation Internet. This paper accordingly proposes an advanced mechanism for this, and gives a differential service model to analyze and simulate the average packet delay for each class. The rest of this paper is organized as follows. The WDM ring network architecture and priority-aware CSMA/CP MAC protocol are presented in Section 2. In Section 3, an approximate non-preemption priority queue model based on an M/G/1 queue with vacations is presented to evaluate the ring performance. In Section 4, the numerical results are obtained from our analysis and simulation. Concluding remarks are made in Section 5.
2. Network architecture
The architecture of a WDM ring network is based on a unidirectional single fiber ring topology; it consists of a number of access nodes (ANs) and W data channels, as shown in Fig. 1. Each AN contains two kinds of network interfaces: (1) the LAN interface is used for the transmission between AN and its access network; (2) the optical link interface is used to access this DWDM MAN ring in the optical domain. Each AN is also equipped with a tunable transmitter and W fixed receivers; each receiver makes use of a particular data channel which has a unique specific wavelength. Every AN can simultaneously receive data from any wavelength, and channels work independently without interference with each other. Logically, the network can be treated as a multi-ring network.
2.1. Structure of the access node
The node structure of the network is shown in Fig. 2. Each node has one tunable transmitter (TT) and W fixed receivers (FRs) dedicated to their particular data channels. For the optical signal sent from upstream nodes, a splitter is used to tap off a small portion of the optical power from the ring to the receivers. Receivers continuously
Fig. 1. Network architecture for the WDM ring network.
Fig. 2. Structure of access node.
monitor sub-carrier headers to detect whether or not the channel is available at that time, and inspect the header information. The data packet will be passed to the local area network (LAN) if its destination address does match this node address. Meanwhile, the MAC control scheme is signaled to activate the semiconductor optical amplifier (SOA) filter to filter the most of the optical signal of the received packet within the delay line interval. When the optical signal arrives at the delay line, it will be delayed for a fixed interval to process the address recognition and adjust the switch array of the SOA Filter to drop the optical signal if necessary. If the destination address of the received packet does not match the node address, the portion signal of the packet in the node will be ignored and most of the delay line will be bypassed to the downstream node. The node then goes back the monitor state. In this network architecture, the destination removal policy is used.
Each node is equipped with multiple fixed receivers, and each takes care of a dedicated data channel; hence the receivers would detect more than one available data channel at a time. However, there is only one tunable transmit- ter in a node to transmit a packet on a specified wavelength at a time; this paper uses the random selection strategy to make the decision for the channel selection. The packets to be transmitted are added into the transmission queue before sending; as the packet is transmitted onto an available data channel, it is propagated in an optical carrier form and its control header is propagated in a sub-carrier frequency multiplexed tone. They are mixed into the optical fiber first, afterward the mixed signal will be sent to the downstream nodes. The number of fixed receivers a node is equipped with is equal to the number of data channels W , and every access node can transmit its data packet to the destination node through any wavelength; therefore, there is no Head of Line (HOL) blocking problem.
Achieving all-optical, DWDM networks will require a simple and effective method to swap headers in real time without affecting the low-latency detection of the data payload. The Sub-carrier Multiplexed (SCM) header tech- nique separates the sub-carrier header from the base-band data payload; it is a relatively easy separation compared to the conventional time domain header techniques. The advantages of using optical labels to encapsulate IP pack- ets are that there is no need to modify the original contents of the IP packet and the header can be coded at the same wavelength of the IP packet. By the SCM scheme, the variable-length IP packets can directly transfer over the DWDM MAN ring networks; it reduces the cost of fragmenting IP packets into many fixed-size time-slots or accumulating IP packets into a large-size frame.
2.2. CSMA/CP MAC protocol
To avoid packet collisions and use bandwidth more efficiently, this paper uses a CSMA/CP MAC protocol [15,16] that is based on carrier sense multiple access and carrier preemption schemes. The carrier sensing scheme is used by the receiver to inspect the sub-carrier signaling of the transmitted packets in an optical fiber. Each wavelength is associated with a sub-carrier frequency; nodes detect the availability of wavelengths by monitoring the sub-carrier in the RF domain [22,23].
Each node monitors wavelengths under the carrier sensing scheme as shown in Fig. 3(a) and tries to find an opening on channels for packet transmission. It is possible that another packet (called a carrier) from upstream
(a)
(b)
Fig. 3. (a) Data payload and its header are sent in wavelength λi. The optical packet arrives and accesses the node and the receiver senses the carrier, which will inform the MAC controller. (b) Delay line will postpone the upstream carrier Tins, the frame will be fragmented at the exact position.
arrives on the same channel when the node is still transmitting its packet, thus a collision occurs. The reason for the collision is the node does not have enough information to know whether the opening on the channel is long enough to accommodate the transmitted packet. Under the carrier preemption scheme, the transmission of a collided packet will be immediately fragmented into two parts: one will be continuously transmitted and the other is added to a queue as shown in Fig. 3(b). The transmitter can continue to transmit the former when the arrival carrier passes into the delay-line. It is noted that the CSMA/CP scheme is done in the electrical domain of the node, but the data packet is transmitted in the all-optical domain of the network. The carrier passes through the delay-line after Tins (nanosecond), just as the transmitter finishes the former transmission. The queued fragment will be transmitted later on the same or another available channel.
To support the carrier preemption scheme, a frame format is designed, as shown in Fig. 4, to solve the addressing capabilities and fragmentation mechanisms. Basically, this consists of a start delimiter (SD), which labels the data frame that is conveyed in the data channel either as packets or fragments. The destinations address (DA) and the source address (SA) fields record the address information in the network. The sequence number (SN) expresses the serial number in a sequence of fragments, and the end fragment (EF) field is used to indicate the last fragment.
Finally, the flag field (FG) is reserved for extended protocol functions, such as defining different service classes for the data payload. The frame length (FL) indicates the length of frame, when the frame is fragmented. Finally, the end delimiter (ED) determines the frame termination. The frame header is composed of SD, DA, SA, SN, EF, FG fields, and the frame trailer is composed of FL, ED fields. To demonstrate the action of packet fragmentation, a collided packet is fragmented into two fragments, as depicted in Fig. 5. The front fragment that has just been transmitted is appended into a frame trailer and the rear fragment for later transmission is inserted into a frame header.
2.3. Priority-aware CSMA/CP MAC protocol
In the current Internet, the same-service-fit-all paradigm, which may cause serious network congestion and packet loss, does not suit real-time multimedia applications, such as video conferencing, Internet phone and video
Fig. 4. The frame format.
Fig. 5. Fragmentation of data frame.
Fig. 6. Concept diagram of priority-aware CSMA/CP MAC protocol.
on demand. In order to support these multimedia applications to operate properly and provide them with the adaptive service-level conditions, the issue of Quality of Service (QoS) has received increasing attention.
The priority-aware CSMA/CP MAC protocol, adding a non-preemption priority function to the transmission queue (TX Queue), is based on the CSMA/CP MAC protocol which has been discussed previously. By marking the Type of Service (TOS) field within the IP packet header, the access node could identify the types of prioritized packet forwarding for each class of service. The concept diagram of priority-aware CSMA/CP MAC protocol is shown in Fig. 6.
3. Performance analysis
The transfer delay of a packet measured from the packet is completely stored in the source node queue until that packet has been completely received by the destination node. This delay consists of queue-waiting delay, transmission delay and propagation delay. The queue-waiting delay of a packet is measured from when a packet is fully stored in a queue of the source node to the time the source node was last selected by the queue before successful transmission. Meanwhile, in this investigation, the transmission delay is defined as the interval between the source node selecting the queue to transmit the packet successfully and the time the source node last selected the queue before transmitting the packet successfully. Finally, the propagation delay of a packet is the interval between the time that the last bit of the packet reaches the destination and the moment that the last bit of the packet was transmitted.
Figure 7 illustrates the timing diagram of a specific node on one channel, considering the ith packet (Pi) arrival into a node transmission queue. This packet must wait in queue for the residual time αiuntil the end of the current
Fig. 7. Calculation of the average waiting time in M/G/1 system with vacations. The average waiting time E[T Qi] of the ith packet is E[T Qi] = E[αi] + E[mi]E[x] + E[Vi].
packet transmission or vacation interval. It must also wait for the transmission of the Mipackets currently in the queue. This queue includes Mi packets, which would be fragmented by upstream traffic as in Fig. 7, and when packet P1 arrivals, it is fragmented into P11 and P12. Finally, the packet must wait during the vacation time Vi because some Mipackets are blocked by upstream traffic. As described above, the expected queue-waiting delay for this packet consists of three items: first, the mean residual time for the packet; second, the expected waiting time for packets ahead of ith packet; and third, the expected vacation times due to blockage by upstream traffic.
From the behavior of the expected queue-waiting delay for the ith packet, the model can be categorized as M/G/1 queue with vacations model [24]. Clearly, the queue-waiting delay captures the effect of contention and is dependent on traffic density. In order to present expressions for packet transfer delay at a node on multi-rings using an M/G/1 vacation model, we first present some assumptions and the general notation to be used in various subsections.
3.1. Assumptions
For simplicity, the following assumptions are made:
1. The number of WDM channels is W .
2. The total propagation delay of the WDM ring is τ seconds, and the distances between the nodes are equal.
3. Packets which arrive are independent, identically distributed (i.i.d.) Poisson process with rate λi (pack- ets/second) at each of the N nodes on the ring, and with an aggregate arrival rate for the network of λ =N−1
i=0 λi.
4. The arrival stream of packets at node i destined for node i⊕ j is a Poisson process with a rate of λi,i⊕j, where⊕ indicates addition modulo N; thus λi=N−1
j=1 λi,i⊕j. In the case of uniform and symmetric traffic on the ring, it indicates that the mean packet generation for all nodes is equal and each source sends equal traffic to all destinations.
λi= λ/N , λi,i⊕j= λi
N− 1 = λ
N (N− 1)
and λi,i= 0, for 0 i N − 1, 1 j N − 1 (1)
5. The packets have random lengths determined at each node as independent, identical and geometrically dis- tributed random variables (denoted by the r.v. M (bits)) with mean E[M ] and probability mass function [25]
Pr(M = k) = β· (1 − β)k, k = 0, 1, 2, . . . where β = 1/(1 + E[M ]).
6. The WDM ring channel bit rate is R (bps) and the packet transmission time without considering vacations is X (= M/R) seconds.
7. Define mTU (minimum transfer unit) as equal to the delay line (L) with Ti = L/R seconds to transmit the mTU.
8. The data packet of Length M would be fragmented into a sequence of nGconsecutive mTUs ignoring the header and trailer length, and assume that Pr(nG= k), k = 0, 1, 2, . . . denotes the probability that nG= k.
Pr(nG= 0) = 0, Pr(nG= 1) =
L M =0
β(1− β)M = 1− (1 − β)L+1,
Pr(nG= k) = Pr
(k− 1)L < M kL
=
1− (1 − β)L
(1− β)(k−1)L+1 and, k = 2, 3, . . . Thus,
E(nG) = [1− (1 − β)L+ (1− β)L+1]
[1− (1 − β)L] (2)
3.2. Notations
The following notations are used in the analytical formulas below:
D average packet transfer delay T Qi queue-waiting delay of packet i
T Q average packet queue-waiting delay
mi number of fragmented packets that must be transmitted before packet i x fragmented packet transmission time
αi residual time of packet i
Vi duration of all the vacation intervals for which packet i must wait before being transmitted V steady-state duration of all the vacation intervals
S average transmission delay
3.3. Analysis of CSMA/CP for single-ring case
With the above assumptions, we model the queue-waiting and transmission delay using an M/G/1 queue with vacations. The average queue-waiting delay, T Qi, for the ith packet is given by
E[T Qi] = E[αi] + E[mi]E[x] + E[Vi] (3)
The queue-waiting delay and transmission delay capture the effect of contention and upstream traffic depen- dence. Thus we consider the delay line (or mTU) as a slot unit, so the dependence is when the full slots are uniformly and independently distributed on a single ring. Our analytical average queue-waiting delay approxima- tions can be obtained by redrawing the timing diagram shown in Fig. 8. Figure 8 illustrates the calculation of the
Fig. 8. Calculation of the average queue-waiting time in a specific node using aggregation of busy time and vacation time. The average waiting time E[T Qi] of the ith packet is E[T Qi] = E[αi] + E[Mi]E[X] + E[Vi].
average queue-waiting time in a specific node using the aggregation of busy time and vacation time. Since the ar- rival process is assumed to be Poisson, this residual time α can be considered to be uniformly distributed between 0 and L/R. Therefore, the mean packet residual time is simply:
E[α] = L
2× R (4)
The number of fragmented packets mi that packet i must wait for is equal to the aggregated number of full packets in the queue. The value of limi→∞E[mi]E[x] is equal to limi→∞E[Mi]E[X] and by Little’s formula, the value of limi→∞E[Mi]E[X] is λiT QE[X]. Letting V = limi→∞E[Vi], we can thus write the steady-state version of Eq. (3):
T Q = E[α] + λiT QE[X] + V (5)
Next we calculate approximate V by multi-channel slotted ring networks. Packets sent by an upstream source use node i as a bridge to reach their destinations, and this bridge has an average traffic load of ρBi =
N−1
k=2
N−1
j=N−k+1λi⊕k,i⊕k⊕jE[Xj].
This upstream traffic blocks the head of the queue packet at node i. Substituting the above assumptions into ρBi gives the following expression:
ρBi=
N−1 k=2
N−1 j=N−k+1
λi⊕k,i⊕k⊕jE[Xj]
=(N− 1)(N − 2)
2 × λ
N (N− 1)×E[M ] R
=(N− 2) × λi× E[M]
2× R
=(N− 2) × λ × E[M]
2× N × R (6)
With this assumption, the average density ρBican be viewed as the probability that mTU is full and continu- ing past the current point. The probability that a packet has to wait i more mTU before it can be transmitted is
ρiBi(1− ρBi). The mean waiting time E[d] to find an empty mTU can be expressed as:
E[d] =
∞ i=0
iL
RρiBi(1− ρBi) = L· ρBi
R(1− ρBi) (7)
When a new packet arrives, it must wait nG· d seconds for each item ahead of it and wait nG· d more for its own service. The steady-state duration of all the vacation intervals V is equal to λiT QE[nG]E[d], and combining Eqs (4) and (7) we obtain the average queue-waiting delay:
T Q = E[α] + λiT QE[X] + λiT QE[nG]E[d]
= L
2· R+ λiT QE[X] + λiT QE[nG] L· ρBi
R· (1 − ρBi) (8)
which can be reduced to:
T Q = L
2· R · (1 − λiE[X]− λiE[nG]R·(1−ρL·ρBi
Bi)) (9)
Because the packet transfer delay is comprised of the queue-waiting delay, transmission delay and propagation delay, the average packet transfer delay is:
D = T Q + S + τ (10)
where τ is the average propagation delay from a source node to a destination node, which is often expressed as τ /2. The average transmission delay is:
S = E[X] + E[nG]E[d]
= E[X] +E[nG]· L · ρBi
R· (1 − ρBi) (11)
Thus, the average transfer delay is given by:
D = T Q + S + τ /2 (12)
3.4. Analysis of CSMA/CP for multi-ring case
In order to analyze the multiple WDM ring networks, it is assumed that the bridge traffic load from the upstream source is equally distributed among W rings. To simplify the analysis, let the circulation of slots on W rings be synchronized [26,27]. That is, a node can observe W mTU on different rings at the same time. Since the bridge traffic load from the upstream source is uniformly distributed among the W rings, the average bridge traffic load of each ring, ρB, can be expressed as:
ρB= ρBi/W (13)
The probability that the packet at the head of a queue cannot get an empty mTU among the currently passing W mTUs is (ρB)W. Therefore, the probability that the packet has to wait i mTUs before it can be sent out is (ρB)W·i(1− (ρB)W).
Fig. 9. A priority-aware queue model of corresponding priority k in node i.
Similar to Subsection 3.3, let E[dB] be the average time required to find the arrival of an empty mTU, then we have:
E[dB] =
∞ i=0
iL
R(ρB)W·i
1− (ρB)W
= L· (ρB)W
R· (1 − (ρB)W) (14)
Since for each packet in the queue the arriving packet has to wait for L/R + dB, the average queue-waiting delay faced by arriving packets is:
T Q = E[α] + λiT QE[X] + λiT QE[nG]E[dB] (15)
Therefore, we have:
T Q = E[α]
1− λiE[X]− λiE[nG]E[dB] (16)
The average transmission delay is:
S = E[X] + E[nG]E[dB]
= E[X] +E[nG]· L · (ρB)W
R· (1 − (ρB)W) (17)
Thus, the average transfer delay is given by:
D = T Q + S + τ /2 (18)
3.5. Analysis of priority-aware CSMA/CP MAC protocol
From the behavior of a priority-aware queue model in Fig. 9, the model can be categorized as a non-preemptive priority M/G/1 queue model [24]. Note that the ρi,kis the product of λi,k(individual class k arrival rate in node i and
λi,k= λi) by E[Xi,k] (the mean service time of class k in node i). The notation γ is the product of E[nG] by E[d].
Consider the mean waiting time Wi,1of the highest priority queue, we obtain:
Wi,1= α + λi,1· Wi,1· E[Xi,1] + λi,1· Wi,1· E[nG]· E[d]
= α
1− λi,1· E[Xi,1]− λi,1· E[nG]· E[d]
= α
1− ρi,1− λi,1· γ (19)
For the second priority queue, we have a similar expression for the mean waiting time Wi,2, except that we have to count the additional waiting time due to packets of higher priority that arrive while a packet is waiting in queue.
Using the expression (19) obtained earlier, we finally have:
Wi,2= α + λi,1· Wi,1· E[Xi,1] + λi,1· Wi,1· E[nG]· E[d]
+ λi,2· Wi,2· E[Xi,2] + λi,2· Wi,2· E[nG]· E[d]
+ λi,1· Wi,2· E[Xi,1] + λi,1· Wi,2· E[nG]· E[d]
= α + λi,1· Wi,1· E[Xi,1] + λi,1· Wi,1· E[nG]· E[d]
1− λi,2· E[Xi,2]− λi,2· E[nG]· E[d] − λi,1· E[Xi,1]− λi,1· E[nG]· E[d]
= α + ρi,1· Wi,1+ λi,1· Wi,1· γ 1− (ρi,1+ ρi,2)− (λi,1+ λi,2)· γ
= α
(1− ρi,1− λi,1· γ)(1 − (ρi,1+ ρi,2)− (λi,1+ λi,2)· γ) (20) The derivation is similar for the general priority queue (class k > 1). The formula for the mean waiting time in queue is:
Wi,k= α
(1−k−1
l=1 ρi,l− [k−1
l=1 λi,l]· γ) · (1 −k
l=1ρi,l− [k
l=1λi,l]· γ) (21)
4. Numerical results
This section presents the simulated and analytical results. The CASI SIMSCRIPT II.5 simulation tool is used to simulate the network model. Here, the behavior of every node is assumed to be the same, and all channels are unidirectional and synchronized in the network. Meanwhile, the packet arrival rate distribution of every node is the same, and the destination of all packets is assigned randomly. Therefore, packets are evenly distributed to all nodes except for their generators. The packet arrival distribution of every node is a Poisson distribution. For a WDM ring with the destination removal policy, each node has one tunable transmitter and W fixed receivers dedicated to their particular data channel. We present some numerical examples to show the correctness of our analyses for average transfer delay. The parameters of the network are shown in Table 1.
Figure 10 presents the simulated and analytical results of average packet transfer delay in this network. The curves demonstrate that a high node offer load can be achieved with low transfer delay when the number of channels is large. Table 2 shows both simulation and analytical results for node offer load where the break points occur in CSMA/CP MAC protocol when the number of channels equals 1, 2, 4 and 8, respectively. When there are 1, 2, 4, and 8 channels, the heaviest offered load (packets/µs) per node is 0.2, 0.5, 1.0 and 2.0. The agreement between the simulation results and the analytical results is excellent.
Table 1 Network parameters
Number of nodes (N ) 16
Number of channels (W ) 1, 2, 4, 8
Ring distance 50 km
Propagation delay of the fiber 5 µs/km
Channel speed OC-192 (10 Gbps)
Size of the delay line 80 ns
Average IP packet size 512 bytes
Fig. 10. Average transfer delay versus offered load per node, when the number of channels equals 1, 2, 4 and 8.
Table 2
Comparison of simulation results and analytical results on average transfer delay versus offered load for CSMA/CP MAC protocol Channel
numbers (W )
Simulation results Analysis results
Offered load Average transfer Offered load Average transfer
(packets/µsec) delay (µsec) (packets/µsec) delay (µsec)
W = 1 0.24 130.2 0.250 127.2
0.25 131.9 0.270 128.2
0.26 136.1 0.276 129.4
0.28 143.7 0.277 131.0
W = 2 0.50 128.3 0.548 127.7
0.53 130.5 0.549 128.6
0.56 135.2 0.550 131.2
0.57 139.9 0.551 143.1
W = 4 0.97 126.6 1.078 127.5
1.10 131.1 1.080 129.0
1.11 133.4 1.081 131.0
1.12 141.3 1.082 141.7
W = 8 1.97 126.6 1.996 127.0
2.00 127.4 2.010 130.3
2.01 128.0 2.012 131.8
2.02 130.3 2.014 138.4
The performance metric using average transfer delay for comparing CSMA/CP and CSMA/CA MAC protocol is shown in Fig. 11. Similar network parameters are found with the CSMA/CA MAC protocol case; under the steady state network condition, the average transfer delay characteristics of the CSMA/CP MAC protocol is better than that of the CSMA/CA MAC protocol. Particularly, the average transfer delay of one channel case in CSMA/CP MAC protocol is almost equal to that of the two channel case in the CSMA/CA MAC protocol. This is coming from the CSMA/CA MAC protocol using the jamming mechanism to handle packet collisions, and the performance would decrease due to the retransmission of the collided packet. This dropped packet wastes the bandwidth resource and decreases the performance of network.
Fig. 11. Comparing CSMA/CP and CSMA/CA: Average transfer delay versus offered load per node, when the numbers of channels equal 1, 2, 4 and 8.
Fig. 12. Average transmission delay under different class (k = 1, 2, 3 and 4) using priority-aware CSMA/CP model.
In the assumption, there are 16 access nodes with attached to four wavelengths (W = 4), and IP packets are divided into four classes. The total traffic is divided equally to each class, i.e., a quarter of the total packet amount.
Figure 12 presents the different performance for each class under the priority-aware model, where Class 4 packets get the worst quality of service and Class 1 packets get the best quality of service. The performance between each service class has a notable diversity. Moreover the agreement between the simulation results and the analytical results is excellent.
5. Conclusions
This investigation describes a novel MAC scheme applicable to all optical WDM ring networks. By facilitat- ing spatial reuse of network bandwidth, the CSMA/CP MAC protocol displays excellent characteristics of high
throughput and low delay for all optical communications. The proposed priority-aware CSMA/CP MAC proto- col adds the class-based priority scheme into the CSMA/CP MAC protocol [15,16]. This novel protocol may differentiate traffic, so when used to implement IP links it is able to help the access nodes implement the quality- of-service-aware (QoS-aware) communication needed in a network that carries multimedia traffic. It also derives the approximate equations for the average packet transfer delay for WDM ring networks with CSMA/CP MAC protocol and priority-aware MAC CSMA/CP protocol respectively. For verification, a simulation program obtains simulated results for the network, and the results closely resemble the analytical values, and this demonstrates the good performance of the network. It is also observed that the throughput characteristic of the network is almost proportional to the number of channels in the network. From simulated results, the throughput of the proposed CSMA/CP MAC protocol has better performance than the CSMA/CA MAC protocol. Transfer delay improves with the number of wavelengths and quality-of–service (QoS) improves with the class-based priority scheme used in the ring, consistent with current WDM technology trends. The characteristic of the ring network is the priority- aware property that introduces the unfairness between nodes, moreover the proposed priority-aware scheme does handle the priority between nodes (i.e., global priority-awareness). Those problems could be left for future re- search.
Acknowledgement
The authors are grateful to the anonymous referees for their help comments. The authors would also like to thank the National Science Council of the Republic of China for financially supporting this research.
References
[1] C.A. Bracket, Dense wavelength division multiplexing networks: principles and applications, IEEE Journal on Selected Areas in Commu- nication SAC-8(6) (1990), 948–964.
[2] P.E. Green, Optical networking update, IEEE Journal on Selected Areas in Communication 14(5) (1996), 746–779.
[3] N. Ghani, S. Dixit and T.S. Wang, On IP-over-WDM integration, IEEE Communication Magazine 38(3) (2000), 72–84.
[4] K. Bengi, Access protocols for an efficient optical packet-switched metropolitan area ring network supporting IP datagrams, Computer Communications and Networks (2002), 284–289.
[5] L. Xu, G. Perros and G. Rouskas, Techniques for optical packet switching and optical burst switching, IEEE Communication Magazine (2001), 136–142.
[6] H. Ogawa, Application of semiconductor optical amplifiers to microwave signal processing, in: IEEE MTT-S International Microwave Symposium Digest, 1995, pp. 1177–1180.
[7] D. Stoll, P. Leisching, H. Bock and A. Richter, Metropolitan DWDM: a dynamically configurable ring for the KomNet field trial in Berlin, IEEE Communications Magazine (2001), 106–113.
[8] J. Cai, A. Fumagalli and I. Chlamtac, The multitoken interarrival time (MTIT) access protocol for supporting variable size packets over WDM ring network, IEEE Journal on Selected Areas in Communication 18(10) (2000), 2094–2104.
[9] S.M. Gemelos, I.M. White, D. Wonglumsom, K. Shrikhande, T. Ono and L. G. Kazovsky, WDM metropolitan area network based on CSMA/CA packet switching, IEEE Photonics Technology Letters 11 (1999), 1512–1514.
[10] K.V. Shrikhande, HORNET: a packet-over-WDM multiple access metropolitan area ring network, IEEE Journal on Selected Areas in Communication 18(10) (2000), 2004–2016.
[11] K. Shrikhande, A. Srivatsa, I.M. White, M.S. Rogge, D. Wonglumsom, S.M. Gemelos and L.G. Kazovsky, CSMA/CA MAC for IP- HORNET: an IP over WDM metropolitan area ring network, in: GLOBECOM 2000, 2000, pp. 1303–1307.
[12] I.M. White, K. Shrikhande, M.S. Rogge, S.M. Gemelos, D. Wonglumsom, G. Desa, Y. Fukashiro and L.G. Kazovsky, Architecture and protocols for HORNET: a novel packet-over-WDM multiple-access MAN, in: GLOBECOM 2000, Vol. 2, 2000, pp. 1298–1302.
[13] M.A. Marsan, An almost optimal MAC protocol for all optical WDM multi-rings with tunable transmitter and fixed receiver, in: ICC’97, 1997, pp. 437–442.
[14] M.A. Marsan, On the performance of topologies and access protocols for high-speed LANs and MANs, Computer Networks and ISDN Systems 26 (1994), 873–893.
[15] W.S. Hwang, W.F. Wang, J.Y. Wang and C.C. Li, A carrier preemption access control protocol for supporting IP packet over WDM ring networks, in: 2001 International Symposium on Communications (ISCOM2001), 2001.
[16] C.C. Li, S.W. Kau and W.S. Hwang, A CSMA/CP MAC protocols for IP over WDM metropolitan area ring networks, in: 8th IEEE International Conference on Communication Systems (ICCS2002), (3P-05-03), 2002, pp. 1217–1221.
[17] W.S. Hwang, J.H. Ho and C.K. Shieh, Performance analysis of IP packets over WDM ring networks, in: 14th IASTED International Conference on Parallel and Distributed Computing and Systems (PDCS2002), 2002, pp. 816–821.
[18] J.H. Ho, W.S. Hwang and C.K. Shieh, Analytical model for an IP over WDM ring network, in: 10th International Conference on Telecom- munication (ICT’2003), 2003, pp. 182–187.
[19] R. Braden, et al., Integrated services in the Internet architecture: An Overview, RFC 1633, Jun. (1994).
[20] S. Blake, et al., An architecture for differentiated services, RFC 2475 (1998).
[21] C. Dovrolis and P. Ramanathan, Proportional differentiated services, Part II: loss rate differentiation and packet dropping, in: Proceedings of IWQoS, 2000, pp. 52–61.
[22] C.L. Lu, D.J.M. Sabido IX, P. Poggiolini, R.T. Hofmeister and L.G. Kazovsky, CORD-a WDMA optical network: subcarrier-based signaling and control scheme, IEEE Photon. Technology Letter 7 (1995).
[23] R. Hui, B. Zhu, R. Huang, C.T. Allen, K.R. Demarest and D. Richards, Subcarrier multiplexing for high-speed optical transmission, IEEE Lightwave Technology Journal 20 (2002), 417–442.
[24] R. Bertsekas and R. Gallager, Data Networks, 2nd edn, Prentice-Hall, Englewood Cliffs, NJ, 1992.
[25] H.M. Ghafir, Performance analysis of a multiple-access ring network, IEEE Transaction on Communications 41(10) (1993), 1494–1506.
[26] C.S. Kang, A broadband ring network: multichannel optical slotted ring, Computer Network and ISDN Systems (1995), 1387–1398.
[27] L.N. Bhuyan, Approximate analysis of single and multiple ring networks, IEEE Transaction on Computers 38(7) (1989), 1027–1040.
![Fig. 7. Calculation of the average waiting time in M/G/1 system with vacations. The average waiting time E[T Q i ] of the ith packet is E[T Q i ] = E[α i ] + E[m i ]E[x] + E[V i ].](https://thumb-ap.123doks.com/thumbv2/9libinfo/9326942.536455/8.892.240.647.188.457/fig-calculation-average-waiting-vacations-average-waiting-packet.webp)
