Sie sind auf Seite 1von 7

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/251937852

Performance comparison of CSMA/CD, CSMA/CA, CSMA/RI, CSMA/PRI and


CSMA/PR with BEB

Article · June 2010


DOI: 10.1109/ICIEA.2010.5515401

CITATION READS
1 1,265

3 authors, including:

P.H.J. Chong wai yie Leong


Auckland University of Technology Taylor's University
235 PUBLICATIONS   2,181 CITATIONS    190 PUBLICATIONS   2,281 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Bio--inspired composites View project

Vehicular Networks View project

All content following this page was uploaded by P.H.J. Chong on 17 May 2016.

The user has requested enhancement of the downloaded file.


Performance Comparison of CSMA/CD, CSMA/CA,
CSMA/RI, CSMA/PRI and CSMA/PR with BEB
Wang Xiaofan, Peter H. J. Chong Leong Wai Yie
School of Electrical & Electronic Engineering Singapore Institute of Manufacturing Technology
Nanyang Technological University, Singapore 71 Nanyang Drive, Singapore 638075
wang0321@ntu.edu.sg, ehjchong@ntu.edu.sg wyleong@simtech.a-star.edu.sg

Abstract—This paper is addressed to analyze the performance Detection (CSMA/CD) with the implementation of the
of various CSMA protocols under disaster scenario. The Truncated Binary Exponential Backoff (BEB) [3]
performance is measured in terms of the total duration to retransmission algorithm [4]. Varieties of MAC protocols have
recover, the mean delay and the throughput when a network been proposed and some are already deployed in practical use.
disaster takes place. Our model includes the widely used BEB The Carrier Sense Multiple Access with Collision Avoidance
algorithm and is implemented for CSMA/CD, CSMA/CA, (CSMA/CA) has been chosen by IEEE 802.11 workgroup for
CSMA/RI, CSMA/PRI, and CSMA/PR in the purpose of the Distributed Coordination Function (DCF) [5] for WLANs
performance comparison. It’s observed that CSMA/PR is able to [6]. In [7], Foh and Zukerman proposed CSMA with
offer satisfying performance for small-size packets, while
reservations by interruptions (CSMA/RI) where stations are
CSMA/RI is better for large-size packets. Also whenever the
network involves priority packets, CSMA/PRI would be an
allowed to interrupt the ongoing transmission and make
optimal solution. reservations of the bandwidth resources. A modified protocol
of CSMA/RI, called CSMA with priority reservations by
Index Terms— total recovery duration, mean delay, interruptions (CSMA/PRI), was proposed in [1]. It categorizes
throughput, CSMA/CD, CSMA/CA, CSMA/RI, CSMA/PRI, all packets for transmission into a few priority grades and only
CSMA/PR allows the stations with the highest priority packets to interrupt
a successful transmission [1].
There have been a lot of performance analyses ever
I. INTRODUCTION since the Ethernet protocol’s appearance in mid 70s [3].
Ethernet has been widely deployed and become the most However, most of them did not take the BEB algorithm into
important LAN technology [1] because the network using account due to complexity. The goal of BEB algorithm is to
Ethernet can carry variable-length IP packets efficiently and is attempt to achieve a 1/ probability of a station
easy to be managed [2]. The Ethernet protocol is based on 1- transmitting on any given transmission opportunity where
persistent Carrier Sense Multiple Access with Collision  is the total number of
stations with packets queued for transmission [8].
Ever since Molle derived the mathematical model for BEB II. PERFORMANCE ANALYSIS
analysis in [9] a lot of works have been done using it to
compare the performance of different MAC protocols. The A. Models, Scenrios and Basic Concepts
comparison of CSMA/RI and CSMA/CD has been done in [4] The performance is analyzed mathematically under the
and in [1] Guo and Kuo also used Molle’s model to analyze the disaster scenario, which models the response of a network to
performance of CSMA/PRI. However, none of them have the recovery from a major failure. It’s termed by IEEE 802.14
made a full comparison of all the similar Carrier Sense to describe a power up situation in the hybrid fiber coax (HFC)
Multiple Access protocols yet. The purpose of this paper is to networks [10]. This situation is likely to occur in local area
analyze various CSMA protocols, i.e. CSMA/CD, CSMA/CA, networks when the shared channel in the network is
CSMA/RI, CSMA/PRI, and CSMA/PR, by Molle’s model temporarily inaccessible due to, for example, a broken cable or
under disaster scenario [10] and evaluate the performance a long period of noise [4].
comparison.
Molle [9] has developed an approximate model for BEB
The rest of this paper is organized as follows. Section II analysis and the accuracy of his work has been verified by
introduces the Molle’s model and various CSMA protocols. simulations. Here we implement the model for various CSMA
And we also show how we apply Molle’s work to analyze protocols and compare the performance in terms of total
different protocols. Sections III shows the result of our duration to recover, the mean delay and the throughput.
numerical results and the comparison of the performance.
Finally, Section IV summarizes the paper. Let ሺሻ denote the idle period, ሺሻ denote the contention
period, and ሺሻ denote the packet transmission period for the
case of  stations. Also ሺሻ is further divided into —ሺሻ
and ™ሺሻ, where —ሺሻ is the period that the channel carries
useful information, and ™ሺሻ is the overhead period.

978-1-4244-5046-6/10/$26.00 2010
c IEEE 1843
Under BEB, the starting time for the next attempt will be , denotes the period of time a network reveries from a disaster
randomly selected from the next 2min(c,10) slots, where counts scenario. And the mean delay, , is the average delay time.
the number of previous attempts for this packet [9]. Thus the
probability that a particular host will make another
transmission attempt at the th time step, is given by [9]: ∑
8
∑ ∑ ∑ ∑
1

where is the probability that a particular host makes its 9


1 st attempt in slot , given that nobody has successfully
acquired the channel during the first 1 steps [9].
∑ ∑ ∑
0 1 2 10

0 3 where ∑ ∑ ∑ ∑ are the


sum of expectations of random variables
1 respectively.
1 10
2
4 B. CSMA/CD
1
1 10 Under CSMA/CD, stations listen to the channel and try to
2 transmit when the channel is idle. The ACK will be send back
upon a successful transmission, otherwise the packets will be
In [9] it is recognized that if the number of active hosts is counted as collisions and be retransmitted. Since packets result
large, we can treat the actions of each of them independently to in collisions are detectable and will be scheduled for
arrive at the average number of transmission attempts in each retransmission, here we count them as contention period to fit
backoff slot. This situation can be represented by the binomial in this model.
distribution 1 and the average number of backoff
slots until the first successful transmission occurs is denoted by 0 11
and is derived in [9] as follows.

1 12
1

13
1 5

14
where

1 1 C. CSMA/CA
1
1 6 The CSMA/CA protocol uses a four-way handshake to
avoid collisions. The use of in-band control frames reserves the
medium for a DATA packet before the packet is transmitted
Hence, the average backoff slots starting from a -host [11]. The basic idea of CSMA/CA is to let both the sender and
disaster scenario is its receiver notify their neighbors before the sender starts to
transmit the DATA. When a node wants to transmit a DATA
packet and finds the channel idle for more than DIFS, it sends
1 1 7 an RTS to the receiver. The receiver sends back a CTS packet
to the sender if the RTS is received correctly. After RTS and
CTS have been exchanged successfully, the transmitter sends
The implementation of this model to different protocols will the DATA packet and receiver returns ACK packet on an error-
be discussed later in this section. Here we define some basic free reception of DATA packet [12].
concepts for performance comparison purpose.
0 15
The throughput, , is defined as the fraction of time that
useful information is carried on the channel. The total duration,

1844 2010 5th IEEE Conference on Industrial Electronics and Applicationsis


1
1 16 0
1 1
24
1
17 1
2 2

0 25
18

D. CSMA/RI 1 0 1
CSMA/RI was proposed in [7] is an extension of the 1-
persistent CSMA/CD by adding reservations by interruptions.
The stations become ready during an ongoing packet 2 1 1
transmission may interrupt the ongoing transmission to make a
reservation of the channel, and then the ongoing packet
transmission is resumed. Upon completion of the ongoing 1 26
transmission, only the stations that have performed reservation
are allowed to access the channel [13].
Due to the reservation by interruptions, not all ready
stations participate during the contention period [4]. We use 27
the same formula as in [4] to denote the which is the
distribution of the number of reservation stations, where is
2 28
the number of stations performed the interruptions among
ready stations.
F. CSMA/PR
1 The packet reservation protocol is similar to the
19
1 1 reservation by interruptions only that it allows stations to make
reservations only during channel idle period.
0 20 0 29

1 1 21 1 30

31
22

32
23

E. CSMA/PRI III. NUMERICAL ANALYSIS RESULTS

In order to further improve the performance of CSMA/RI, a In this section, we present the numerical simulation results
priority reservation by interruptions protocol was proposed in and compare the performance of different protocols. We run
[1] where stations need to go through two phases of the numerical computation for two sizes of data packets, 10
reservations, called first reservation and second reservation time slots and 25 time slots respectively, to see how the
respectively, to access the channel. The packets are classified protocols’ performance would be affected. The parameters are
into three classes, and only the stations with the lowest priority defined as in Table.1. We assume the stations are able to
packets are allowed to make the second reservation directly. detect idle channel immediately upon completion of a
We use the similar model for CSMA/PRI only that we successful transmission and reservation phase takes only one
simplify the packets to only two classes, denoted by . Packets time slot. For the CSMA/PRI protocol, we assume the
with =0 are high-priority packets and they perform reservation proportion of is
by interruptions in the same way as CSMA/RI, only those
packets with =1 need to make two reservations. TABLE I. PARAMETERS FOR NUMERICAL ANALYSIS

2010 5th IEEE Conference on Industrial Electronics and Applicationsis 1845


Channel Bit Rate 10Mbit/s In terms of the total duration and mean delay, it’s observed
Slot Time 50μsec that CSMA/CA outperforms CSMA/CD by exchanging
DIFS Two Slots RTS/CTS, however the performance is still worse off than
SIFS One Slot reservation based protocols, i.e. CSMA/RI, CSMA/PRI and
RTS Packet Size One Slot CSMA/PR. The latter three achieve similar performance and
CTS Packet Size One Slot more details are investigated below.
ACK Packet Size One Slot
0.8
DATA Packet Size (i) 10 Slots
(ii) 25 Slots
0.7

0.6
A. Small Packet Size (10 time slots)
--- o --- CSMA/CA
--- * --- CSMA/CD

Throughput
4000 0.5
--- + --- CSMA/RI
--- o --- CSMA/CA --- x --- CSMA/PRI
3500 --- * --- CSMA/CD 0.4 ---------- CSMA/PR
--- + --- CSMA/RI
--- x --- CSMA/PRI
3000 ---------- CSMA/PR 0.3
Total Duration (Ttotal)

2500
0.2

2000
0.1
10 15 20 25 30 35 40 45 50 55 60
1500 Number of Users (M)

1000 Figure 3. Throughput (data packet size = 10 slots)

500 The three protocols, CSMA/RI, CSMA/PRI and


CSMA/PR, are able to achieve much higher throughput
0 because their reservation phase consumes only one time slot
10 15 20 25 30 35 40 45 50 55 60
Number of Users (M) while the other two have very significant overhead. However,
our interest is to analyze the performance comparison among
Figure 1. Total recovery duration (data packet size = 10 slots) these three to see which one is more optimal and applicable.
In order to do that, we look at the throughput of the three
protocols of our interest and also the performance of just high
priority packets in CSMA/PRI.
60
0.8
--- + --- CSMA/RI
--- x --- CSMA/PRI
50 ----------CSMA/PR
--- o --- CSMA/PRI for packets with high piority
0.75

40 --- o --- CSMA/CA


Mean Delay (Davg)

--- * --- CSMA/CD


--- + --- CSMA/RI 0.7
Throughput

--- x --- CSMA/PRI


30 ---------- CSMA/PR

0.65
20

10 0.6

0
10 15 20 25 30 35 40 45 50 55 60 0.55
Number of Users (M) 10 15 20 25 30 35 40 45 50 55 60
Number of Users (M)

Figure 2. Mean delay (data packet size = 10 slots) Figure 4. Throughput (data packet size = 10 slots)

1846 2010 5th IEEE Conference on Industrial Electronics and Applicationsis


As shown in Fig. 4, although the performance of
CSMA/PRI is worse off than CSMA/RI as a whole, the 140

throughput of high priority packets is higher than any other


protocol. Therefore when the traffic nature requires ensuring 120
successfully transmission of certain high-priority packets,
CSMA/PRI would be the optimal solution to such scenario. 100
However, it’s also noted that CSMA/PR achieves very similar --- o --- CSMA/CA

Mean Delay (Davg)


--- * --- CSMA/CD
performance to CSMA/RI, but the complexity of the protocol 80 --- + --- CSMA/RI
implementation is much lower. Hence CSMA/RI can be a --- x --- CSMA/PRI
---------- CSMA/PR
promising protocol for small-size packets networks when the
60
priority issue is not involved.
40
B. Large Packet Size (25 time slots)
Next we run the same computation model for larger packet 20
size, 25 time slots, to investigate how the performance is
affected. Also it’s of our interest to see whether each protocol 0
10 15 20 25 30 35 40 45 50 55 60
is still able to offer same performance when the packet size Number of Users (M)
varies.
As expected the total recovery duration increase almost Figure 6. Mean delay (data packet size = 25 slots)
proportional to the data packet size. On the contrary, the
throughput increases with the data size because the overhead
1
packet size does not change. But the change of mean delay is
not significant for most of the protocols except CSMA/CD, 0.9
because large-size packets cause the increase of overhead for
CSMA/CD. 0.8
From Fig. 5, 6, and 7 we observe that the performance of
0.7
CSMA/CA is still better than CSMA/CD, while worse off the
other three. CSMA/RI, CSMA/PRI and CSMA/PR still
Throughput

0.6
achieve similar total recovery duration and mean delay.
However, the relationship among their throughput has changed 0.5
for larger packet size scenario and this issue is discussed --- o --- CSMA/CA
below. 0.4 --- * --- CSMA/CD
--- + --- CSMA/RI
0.3 --- x --- CSMA/PRI
9000 ---------- CSMA/PR
--- o --- CSMA/CA 0.2
8000 --- * --- CSMA/CD
--- + --- CSMA/RI
0.1
7000 --- x --- CSMA/PRI 10 15 20 25 30 35 40 45 50 55 60
---------- CSMA/PR
Total Duration (Ttotal)

6000
Figure 7. Throughput (data packet size = 25 slots)
5000
When the size of data packets increases, the CSMA/RI and
4000 CSMA/PRI protocols achieve much higher throughput than
CSMA/PR although the performance of them were similar
3000
when a smaller packet size was used. This is because the
2000 CSMA/PR only allows stations to make reservations during
idle period, which is very short when the packet size is large.
1000 On the other hand, no matter how large the packet size is,
CSMA/RI and CSMA/PRI are always able to make
0
10 15 20 25 30 35 40 45 50 55 60 reservations because they may interrupt the ongoing
Number of Users (M) transmission.
Figure 5. Total recovery duration (data packet size = 25 slots) Hence CSMA/PRI and CSMA/RI may be a better choice
when the packet size is large although their complexity is
higher. Especially when priority packets are involved in the
networks, using CSMA/PRI is able to offer a throughput
higher than 90% as shown in Fig. 8.

2010 5th IEEE Conference on Industrial Electronics and Applicationsis 1847


0.94
--- + --- CSMA/RI
--- x --- CSMA/PRI IV. CONCLUSION
----------CSMA/PR
0.92 --- o --- CSMA/PRI for packets with high piority
In this paper, we used the mathematically model for BEB
derived by Molle [9] to compare the performance of
CSMA/CD, CSMA/CA, CSMA/RI, CSMA/PRI, CSMA/PR
0.9 under disaster scenario. The comparison was done for two
different packet sizes and we conclude that CSMA/PR is more
Throughput

0.88
suitable for small packet size while CSMA/RI can achieve
better performance for large packet size. Also it’s noted that
whenever the networks have special requirement to ensure
0.86 transmission for high priority packets, CSMA/PRI is always
able to offer satisfying performance.
0.84

REFERENCES
0.82 [1] H. B. Guo and G. S. Kuo, "CSMA with Priority Reservation by
10 15 20 25 30 35 40 45 50 55 60 Interruptions for Efficiency Improvement and QoS Support," Consumer
Number of Users (M) Communications and Networking Conference, pp. 456 - 460, 2005.
Figure 8. Throughput (data packet size = 25 slots) [2] G. Kramer and G. Pesavento, "Ethernet Passive Optical Network
(EPON): Building A Next-Generation Optical Access Network," IEEE
Communication Magazine, vol. 40, pp. 66 -73, Feb 2002.
[3] R.M.Metcalfe and D.R.Boggs, "Ethernet: Distributed Packet Switching
C. Performance with Different Packet Sizes for Local Computer Networks," Communications of the ACM, vol. 19,
July 1976.
As discussed above, the CSMA/CD and CSMA/CA cannot [4] C.H.Foh and M. Zukerman, "Performance Comaprison of CSMA/RI and
achieve high performance as those of reservation based CSMA/CD with BEB," Proc. of ICC, vol. 9, pp. 2670 - 2675, June 2001.
protocols due to significant overhead. The other three [5] P802.11, "IEEE Standard for Wireless LAN Medium Access Control
methods, CSMA/RI, CSMA/PRI and CAMS/PR, benefit a lot (MAC) and Physical Layer (PHY) Specifications," Nov 1997.
from the reservation phase, although adding on complexity to [6] C. Heng and M. Zukerman, "Performance Evaluation of IEEE 802.11,"
IEEE Vehicular Technology, vol. 57, pp. 3757 - 3768, Nov 2008.
the implementation.
[7] C.H.Foh and M.Zukerman, "CSMA with reservations by interruptions
(CSMA/RI): A novel approach to reduce collisions in CSMA.CD," IEEE
TABLE II. PERFORMANCE COMPARISON FOR DIFFERENT PACKET SIZES
Journal on Selected Areas in Communications, vol. 18, pp. 1572 -1580,
Sep 2000.
Small Data Packet Large Data Packet
[8] K. J. Christensen, "Performance Evaluation of the Binary Logarithmic
CSMA/RI It can achieve When the data packet is Arbitration Method (BLAM)," Proc. Local Computer Networks, pp. 396
similar performance large, CSMA/RI is the - 403, Oct 1996.
as CSMA/PR does, optimal protocol. [9] M. Molle, "A New Binary Logarithmic Arbitration Method for
but with more Ethernet," Technical Report CSRI- 298, Computer systems Reserch
complexity. Institute, University of Toronto, Canada, 1994.
[10] C.Bisdikianm, "A review of random access algorithm," IEEE 802.14
CSMA/PR CSMA/PR is the The throughput of Working Group Document No. 802.14 - 96/019, Jan 1996.
optimal protocol for CSMA/PR is worse [11] J. Peng and L. Cheng, "Revisting Carrier Sense Multiple Access with
small data packets than the other two Collision Avoidance," Proc. Information, Sciences and Systems Conf.,
because it’s easy to because the idle period pp. 1236 - 1241, March 2006.
implement and which reservations can [12] A. Tsertou and D. I. Laurenson, "Revisting the Hidden Terminal
achieves similar take place is much Problem in a CSMA/CA Wireless Network," IEEE Transcations on
Mobile Computing, vol. 7, pp. 817 - 831, July 2008.
performance as shorter.
[13] C.H.Foh and M.Zukerman, "Improving the Efficiency of CSMA using
CSMA/RI. Reservation by Interruptions," Proc. IEEE ICC, June 2000.
CSMA/PRI When the network needs to transmit packets
with different priority levels, CSMA/PRI can
always ensure very high performance for those
high-priority packets.

1848 2010 5th IEEE Conference on Industrial Electronics and Applicationsis

View publication stats

Das könnte Ihnen auch gefallen