Authorized licensed use limited to: St. Josephs College of Engineering. Downloaded on July 22,2010 at 03:22:09 UTC from IEEE Xplore. Restrictions apply.
The rest of this paper i s organi zed as follows.
Section 2 presents static inter-arri val time. Section 3
presents d ynamic inter-arri val time. Secti on 4
presents simulation results and anal ysis that probe
the efficiency of our proposed scheme. And final
section concludes this work.
892
Authorized licensed use limited to: St. Josephs College of Engineering. Downloaded on July 22,2010 at 03:22:09 UTC from IEEE Xplore. Restrictions apply.
Inter-arrival time flow control is desirable to be as small as possible. Fig5 shows a
Inter-arrival time flow control is the simplest flow FCRP format. Source, destination address virtual
control technique. Fig4 represents inter-arrival time path identifier have 8 bit respectivel y.
flow control algorithm. In S-IAT OBS network, each
data burst is sent at allocated ti me interval.
∑ (D
Consequentl y all edge nodes work in order to keep (2)
the data burst contention rate in the acceptable area. O + TO )
Co is calculated with simple moving average.
3. Dynamic inter-arrival time flow control Si mple moving average can interpret accurate data
The second proposed scheme is Dynami c
burst contention rate regardless burst y traffic due to
inter-arrival time flow control (D-IAT) which aims
applying a constant t erm. Fig 6 shows the idea of
to achieve higher throughput than S-IAT. The basi c
moving average. If Ǎ t is longer than a proper one,
idea of D-IAT is same as S-IAT. In D-IAT OBS
then a response speed of moving average for
network, core node request edge nodes to decrease
calculation of data burst contention rate will be
traffi c which gets through a congested link when it is
impervious and consequently QoS i n terms of data
detect ed. Each edge node which received a flow
burst contention rate will be deteriorated. On the
control request packet decreases or keeps target VP’s
other hand, in a case that Ǎ t is too short, a
throughputs so that every edge nodes keep fair share.
congestion state will be detected too sensitivel y and
D-IAT is composed of t hree el ements, Flow Control
consequently FCRPs will be sent wastefull y. Note
Request Packet(FCRP), D-IAT core node, and D-IAT
that Ǎ t affects the accuracy of data burst contention
edge node.
rate. D-IAT core nodes interpret its output link as
congestion state when the data burst CO exceeds the
Flow Control Request Packet(FCRP)
data burst contention rate threshold. Thi s calculation
In D-IAT OBS, D-IAT core nodes create Flow
is tri ggered whenever a data burst i s dropped.
Control Request Packets (FCRP) to request D-IAT
The second step of D-IAT core node function is to
edge nodes of flow control. FCRP is sent on a
determine to send FCRPs. D-IAT core nodes need to
control channel same as control packets. Therefore it
893
Authorized licensed use limited to: St. Josephs College of Engineering. Downloaded on July 22,2010 at 03:22:09 UTC from IEEE Xplore. Restrictions apply.
wait for FCRP’s effect s until they reach their inter-arrival time, the congested output link will be
destination D-IAT edge nodes due to propagation worse and dat a bursts will be increasingl y dropped.
delay. In case of Fi g 6 for example, Co exceeds the In order to sol ve this problem, D-IAT edge node puts
data burst contenti on rate threshold and D-IAT core current inter-arri val times back to ori ginally ones
node send FCRPs at t0. However Co keeps which are used in -IAT. ote that the D-IAT edge
increasing until they reach their destination edge node increases the only inter-arri val time on
nodes. If FCRPs are sent regardless of their congested VPs. Conse uently other non-congest ed
propagation delay, D-IAT edge nodes will decrease VPs have been affected.
their flow rate even though they do not need. In
order t o avoid this situation, D-IAT core node waits 4. Simulation Results and Analysis
to send FCRPs for the maximum round trip t ime In this study, we perform a number of simulations
(RTT). The maximu m RTT is calculated from VPs on a mesh network in order to evaluate the
which get through this congested output link. performance of the proposed congestion avoidance
scheme. In this simulation, we consider a NSFNET
topology with 14 nodes as shown in Fig 7. In this
model, it is assumed that a single fi ber has 16
channels and each channel is operating at 2.5Gbps. A
static virtual path is chosen between each pair of
nodes. The switching time and the processing time of
a control packet in each node are set to 5Ǵ s. Data
bursts are generated according to a Markov Modulate
Poisson Process (MMPP) and have a fixed size
(100Kbit). Simulation time i s set to 500,00,00Ǵ s.
Also it is assumed that no intermediate buffers, no
wavelength conversion and no deflection routing are
used in the nodes. In this study, the data burst
Fig6 simple moving average contention rate is set to 0.05.
894
Authorized licensed use limited to: St. Josephs College of Engineering. Downloaded on July 22,2010 at 03:22:09 UTC from IEEE Xplore. Restrictions apply.
㪊㪅㪌㪌㪜㪂㪇㪏
0.2
㪊㪅㪌㪇㪜㪂㪇㪏 Conventional
㪊㪅㪋㪌㪜㪂㪇㪏 S-IAT
㪊㪅㪊㪌㪜㪂㪇㪏
㪊㪅㪊㪇㪜㪂㪇㪏 0.1
㪊㪅㪉㪌㪜㪂㪇㪏
㪊㪅㪉㪇㪜㪂㪇㪏
0.05
㪊㪅㪈㪌㪜㪂㪇㪏
㪊㪅㪈㪇㪜㪂㪇㪏
㪊㪅㪇㪌㪜㪂㪇㪏 0
1.25E-03 6.25E-03 1.13E-02
㪉㪌㪇㪇㪇㪇 㪉㪌㪉㪇㪇㪇 㪉㪌㪋㪇㪇㪇 㪉㪌㪍㪇㪇㪇 㪉㪌㪏㪇㪇㪇 㪉㪍㪇㪇㪇㪇
Data Burst Generation Rate Lambda
㪪㪸㫄㫇㫃㫀㫅㪾㩷㪫㫀㫄㪼㩿㱘㫊㪀
Fig8 throughput during 25000Ǵ s and 26000Ǵ s Fig9 Characteristics Data burst contention rate
In the second simulation, we observe the dat a Finally, in the forth simulat ion, we observe the deliver y
burst contention rate characteristic. Fig 9 shows the delay character istic. Fig 11 shows the average delivery
data burst contention rate of one VP as function of delay o n o ne VP as function of the data burst generation
data burst generation rate λ . The data burst rate λ . The delivery delay of the conventional scheme
contention rate of con ventional scheme keeps higher keeps a constant rate since it does not suffer flow control
as λ gets higher. However the data burst at edge nodes. On the other hand, S-IAT suffers the
contention rates of both proposed schemes keep the longest delay in these schemes because their inter-arrival
critical data burst contention rate, which is the upper time is fixed. Due to fixed inter-arrival time, Fig.10
limit of the acceptable area. We observe that the data shows that S-IAT wastes the network resources. D-IAT
burst contention rate of D-IAT is slight higher than which supports a dynamic inter-arrival time can achieve
S-IAT. shorter delivery delay than S-IAT one. This is because
In t he third simulation, we observe the throughput D-IAT edge nodes cha nge their inter-arrival times and
characteristic. Fig 10 shows the throughput as send data bursts according to the network congestion state.
function of dat a burst generation rate λ of global D-IAT makes it possible to utilize network resources more
network. X-axis and Y-axis mean data burst effectively than S-IAT.
generation rate λ and throughput respecti vel y. Th e
throughput of S-IAT is increasing until λ reaches 0.1
to 0.00625. Thi s is because when λ is below S-IAT
0.00625, the load is not the critical l oad. However 0.08 D-IAT
the throughput of S-IAT keep a constant value when
Throughput(Gbps)
λ is over the critical l oad for keeping the data burst 0.06
contention rate in the acceptable area. On the other
hand, the throughput of D-IAT is increasing even 0.04
though λ exceeds the critical load. This is because
D-IAT edge nodes send the many more data bursts 0.02
than S-IAT so that the network resources are used
effectivel y when network is non-congestion state. 0
D-IAT achieve 35% higher throughput than S-IAT. 1.25E-03 6.25E-03 1.13E-02
Data Burst Generation Rate Lambda
895
Authorized licensed use limited to: St. Josephs College of Engineering. Downloaded on July 22,2010 at 03:22:09 UTC from IEEE Xplore. Restrictions apply.
“Photonic Packet Switches: Architect ures and
㪋㪈㪋㪌 Exper imental Implementat ions”, Proceedings of the
㪚㫆㫅㫍㪼㫅㫋㫀㫆㫅㪸㫃 IEEE, vol. 82, pp. 1650-1667, Nov. 1994.
[4] L. Dittmann, C. Develder, D. Chiaroni, F. Neri, F.
㪋㪈㪋㪇 㪪㵥㪠㪘㪫 Callegati, “The European IST Project DAV ID: A
Viable Approach Toward Optical Packet Switching”,
㪛㪼㫃㫀㫍㪼㫉㫐㩷㪛㪼㫃㪸㫐㩿 㱘㫊㪼㪺㪀
㪛㵥㪠㪘㪫
IEEE Journal o n Selected Areas in Communications,
㪋㪈㪊㪌 VOL. 21, NO. 7, SEPT 2003.
[5] Christian Guillemot, Monique Renaud, Piero
Gambini “Transparent Optical Packet Switching: The
㪋㪈㪊㪇 European ACTS KEOPS Project Approach”, Jour nal
of Lightwave Technology, VOL. 16, NO. 12 DEC
1998.
㪋㪈㪉㪌
[6] Chunming Qiao, Myungsik Yoo, “Optical burst
switching (OBS) - a new paradigm for an optical
㪋㪈㪉㪇 Internet”, Journal of High Speed Networks, VOL. 8,
Issue. 1, March 1999.
㪈㪅㪉㪌㪜㪄㪇㪊 㪍㪅㪉㪌㪜㪄㪇㪊 㪈㪅㪈㪊㪜㪄㪇㪉
[7] Yijun Xiong, Marc Vandenhoute, Hakki C. Cankaya,
㪛㪸㫋㪸㩷㪙㫌㫉㫊㫋㩷㪞㪼㫅㪼㫉㪸㫋㫀㫆㫅㩷㪩㪸㫋㪼㩷㪣㪸㫄㪹㪻㪸 “Control Architecture in Optical Burst-Switched
WD M Networks”, IEEE Journal on Selected Areas i n
Communications, Vol. 18, No. 10, OCT 2000.
Fig 11 Trans mission Delay
[8] Barakat, N. Sargent, E. H. “D ual-header opt ical burst
switching: a new architect ure for WDM
5. Conclusion burst-switched networks” INFOCOM 2005. 24th
Annual Joint Conference of the IEEE Computer and
In Opti cal Burst Switching (OBS), edge nodes Communications Societies. Proceedings IEEE, VOL.
keep sending data bursts regardl ess of the network 1, pp 685-693.
condition which may overwhelm the network [9] Sungcha n Kim, “Contention Resolution for Optical
Burst Switching Networks Using Alternative
capacities. It causes the serious burst contention Routing”, 2002 IEEE International Conference o n
problem in OBS. In order to sol ve this problem, we Communications, vol.101,no.2,May 2002
have proposed two novel schemes named S-IAT and [10] Yoshihiko Mori etc, ”Effective Flow rate
Control for the Deflection Rout ing based Optical
D-IAT based on the VP concept. They allow that Burst Switching Networks,” Asia-Pacific Conference
edge nodes send the optimum traffic flow for the on Communications 2006.
OBS network. S-IAT allows keeping the dat a burst [11] C.Gauger, “Dimensio ning of FDL Buffers for
Optical B urst Switching Nodes,” Proceedings,
contention rate in the acceptable state. D-IAT allows Optical Network Design and Modeling(ONDM)2002,
detecting congestion state with simple moving Torino, Italy, 2002
average and virtual path. D-IAT has achieved not [12] Jeyashanker Ramamirtham Turner, J. , “Time
sliced optical burst sw itching”, INFOCOM 2003.
only keeping the data burst contention rate in Twe nt y-Second Annual Joint Confere nce of the IEEE
acceptable area but obtaining 35% higher throughput Computer and Communications Societies. IEEE, Vol.
and lower delivery delay than S-IAT. 3, pp. 2030-2038.
[13] D.E.McDysan, D.L.Spohn , “ATM - Teory and
Application”, McGraw-Hill Inc, 1994
ACKNOWLEDGMENT [14] Abdelilah Maach, Gregor V. Bochman, Hussein
"This research was partly supported by both the Mouftah, “Congestion Control and Contentio n
Elimination in Optical Burst Switching”,
Ministry of Knowledge and Economy, Republic of Telecommunicat ion System Vol. 27. pp. 115-131
Korea, under the ITRC support program supervised 2004.
by IITA(IITA-2009-C1090-0902-0004), and the
Ministry of Education, Science, Sports and Culture
Japan, Grant-in-Aid for Scientific Research (B), no.
19360180, 2009.
References
[1] Alexander Birman, “Computing Approximate
Blocking Probabilities for a Class of All-Optical
Networks”, IEEE Jour nal on Selected Areas i n
Communications, VOL. 14, NO. 5 JUNE 1996.
[2] E. Karasan and E. Ayanoglu, “Effects of Wavele ngth
Routing and Selection Algorithms on Wavelengt h
Conversion Gain in WDM Optical Networks,”
IEEE/ACM Trans. Networking, 6, pp. 186-196, 1998.
[3] D. J. Blumenthal, P. R. Prucnal, and J. R. Sauer,
896
Authorized licensed use limited to: St. Josephs College of Engineering. Downloaded on July 22,2010 at 03:22:09 UTC from IEEE Xplore. Restrictions apply.