Sunteți pe pagina 1din 5

MMSRP: Multi-wavelength Markov-based Split Reservation

Protocol for DWDM Optical Networks


Malabika Sengupta Swapan Kumar Mondal Debashis Saha
Kalyani Govt. Engg. College Kalyani Govt. Engg. College IIM Calcutta
Kalyani, West Bengal, India. Kalyani, West Bengal, India. Kolkata, India.
913325288537 913325941794 913324622833
ms5_1@yahoo.com paltaswapan@yahoo.co.in ds@iimcal.ac.in

ABSTRACT Keywords
During wavelength reservation in Wavelength Division Optical networks, WDM, splitting, Markov model
Multiplexed (WDM) optical networks, often multiple
connection requests unknowingly compete for the same 1. INTRODUCTION
wavelength, even when other free wavelengths are available, In Dense Wavelength Division Multiplexed (DWDM) optical
resulting in a collision. The Markov model used in Markov networks [1]-[2],[4] without wavelength converters, a dedicated
Based Reservation Protocol (MBRP), is very effective to reduce lightpath [5] is first established between the source-destination
such conflicts by intelligently guessing a wavelength in advance. pair of a connection request (henceforth we refer request to
Even then a connection request may be blocked because of the represent connection request) before the actual data transfer
vulnerable period between wavelength probing and actual starts. Lightpath establishment involves three basic steps: (i)
reservation. To minimize the effect of such vulnerability, routing, (ii) wavelength selection, and (iii) wavelength
splitting the probe process to fork out a partial reservation from reservation [1], [3]-[5]. This work is focused on wavelength
an intermediate node is an efficient solution. In Markov- selection and reservation. Here we have used fixed shortest path
selection Split Reservation Protocol (MSRP), the above two routing, but any other routing may be equally applicable.
strategies are combined, but only one wavelength is guessed
during probing. If the attempt with this single wavelength fails, A judicious selection of wavelength from an available set of
the connection request is blocked. To take care of this limitation, wavelengths, reduces the probability of collisions among the
we propose here a new scheme called Multi-wavelength MSRP concurrent requests. Random-fit and first-fit [2] are two
(MMSRP), where a set of wavelengths (instead of one) is
conventional methods for wavelength selection, whereas the
selected by Markov model and continuously updated for
recently proposed methods include label prioritization [3], and
possible future use. In case of failure, during reservation in the
backward direction, it retries to reserve the next best wavelength Markov selection [8],[10]. In label prioritization, the priorities
through another splitting at the failure point. Thus, MMSRP of wavelengths are set depending on the duration of stay in the
handles multiple wavelengths sequentially through multiple pool. In Markov method, a particular wavelength is guessed
splitting. Simulation results show that the blocking probability from the available pool of wavelengths well in advance, for each
in MMSRP decreases considerably (~25% over MSRP and request separately. Thus, other concurrent requests can exclude
~50% over MBRP in some cases) as the number of wavelengths the already guessed wavelengths and consider the rest, thereby
increases. Compared to MSRP, though the average setup time is reducing the probability of wavelength conflict. We use this
marginally higher in MMSRP, it appears quite promising for strategy for selection of wavelength in this work.
delay-tolerant applications, where blocking is very crucial, in
dense WDM networks. Successful reservation of the selected wavelength throughout the
path, is also very crucial because information about wavelength
Categories and Subject Descriptors availability is difficult to be guaranteed at any particular place
B.4.4 [Input/Output and Data Communications]: and time in large DWDM networks. To handle this issue, several
Performance Analysis and Design Aids – Formal models, reservation protocols [3], such as Source Initiated Reservation
Simulation. Protocol (SIRP), Destination Initiated Reservation Protocol
(DIRP), Intermediate node Initiated Reservation Protocol (IIRP)
General Terms [6] and Split Reservation Protocol (SRP) [7] have been reported
Performance. in literature. Two recent additions to this list are: Markov Based
Reservation Protocol (MBRP) [10], and Markov-selection Split
Permission to make digital or hard copies of all or part of this Reservation Protocol (MSRP) [8]. In MBRP, selection of
work for personal or classroom use is granted without fee wavelength is done using Markov method and reservation is
provided that copies are not made or distributed for profit or done using DIRP. Here, a wavelength is guessed for a particular
commercial advantage, and that copies bear this notice and the request well in advance, so that other requests do not select that
full citation on the first page. To copy otherwise, to republish, to guessed wavelength. Thus, wavelength conflict among
post on servers or to redistribute to lists, requires prior specific contemporary requests is reduced. Consequently, MBRP
permission and/or a fee. performs better than DIRP [10]. In MSRP, wavelength selection
SAC’10, March 22-26, 2010, Sierre, Switzerland. is done using Markov model and reservation of wavelength is
Copyright 2010 ACM 978-1-60558-638-0/10/03…$10.00. done using SRP. Using the concept of splitting [7], blocking is
reduced by shortening the vulnerable period. So MSRP performs
better than MBRP [8]. However, a major limitation of MSRP is represents that the corresponding wavelength is free and a “0”
that, it attempts only one wavelength for reservation and otherwise, future_guess_set is an array which contains a set of
splitting is done only once. If the selected wavelength fails guessed wavelengths for probable use in future if the request is
during reservation, the request is blocked. Hence, there is still splitted, prev_guess_index contains the guessed wavelength upto
space for further improvement by extending Markov selection to the previous node. Use of future_guess_set and
multiple wavelengths and subsequently incorporating multiple prev_guess_index are discussed in Subsection 2.2.
splitting. This concept helps many erstwhile failure cases
succeed, thereby reducing the overall blocking of requests. We While PROB moves towards destination, each node performs
call this proposed scheme as MMSRP. three major tasks discussed in the following subsections.
The paper is organized as follows. Wavelength selection in
MMSRP is described in Sections 2 whereas reservation of 2.1 Detection of interfering requests
selected wavelength(s) is discussed in Section 3. Comparative When a request arrives at a node, it is called current request. All
simulation results are discussed in Section 4. Finally, Section 5 other ongoing requests that arrived earlier at that node are called
concludes the paper. under process requests. Those under process requests, whose
pre_hop_id or next_hop_id (available in node_table) matches
with next_hop_id of the current request, are called interfering
2. WAVELENGTH SELECTION IN requests. When a PROB arrives at a node, the node detects the
MMSRP interfering requests using node_table. This detection of
Each node of the network maintains three tables: node_table, interfering requests is required to identify and exclude the
link_status_table and markov_table. A node_table keeps records wavelengths guessed by them to reduce the probability of
of all requests passing through the node. Each record of node_ collision of wavelengths.
table contains a few fields. All these fields are mentioned below.
Connection_id: identity number of the request, source_id: 2.2 Selection of the guessed wavelength
identity number of source of a route (say R), destination_id: We assume that each node broadcasts its adjoining link usage
identity number of destination of R, pre_hop_id: identity information at every T seconds. This is stored in
number of previous node in R, next_hop_id: identity number of link_status_table. So each node comes to know about the
next node in R, arrival_time: time when the request arrives at wavelength allocation on all links in the network at times 0T,
the present node and guessed_wavelength: the wavelength 1T, …, sT, …, where s is a natural number. However, the
guessed in the present link by the present node. The duration of information broadcast at sT may not remain correct at an
a record in a node_table is bounded by the estimated source- arbitrary time instant between sT and (s+1)T. To overcome this
destination round trip time of the concerned request. uncertainty, a prediction based on C-T Markov chain [6] is used.
Markov parameters are broadcast at every T ′ seconds and stored
We assume that each node broadcasts its adjoining link usage in markov_table of each node. Usually, T ′ >> T so that the
information at every T seconds. This link usage information is Markov chain remains stable. After receiving a PROB, each
stored in link_status_table at every node. The records in node first updates its probe_map field by marking those
link_status_table contains the fields: link_id and bit_map. The wavelengths of the present link (the link connecting the present
field bit_map is of size equal to the number of wavelengths used node and the next node of the route) as busy (if any), which are
in the network and represents status of usage of all wavelengths (i) already guessed by interfering requests or (ii) reserved by
of the link. A ‘1’ is placed in the bit_ map when the other requests for future transmission or (iii) reserved and being
corresponding wavelength is free and a ‘0’ otherwise. used for transmission. If no free wavelength is available, PROB
fails and the request is blocked. A NACK is generated and sent
The markov_table contains the information of rate of change of towards source. NACK releases the entries of this request from
states of the wavelength usage for all the wavelengths in all the node_table at each node till it reaches source. Otherwise, for
links. The records of markov_table contains link_id and each free wavelength (if any), the node uses the markov_table to
rate_map. The rate_map contains rate of change of states of the find the probability of getting the wavelength free throughout
wavelength usage for all the wavelengths in the corresponding the path [10]. We extend the concept of guessing [10], using
link_id. multiple number of wavelengths. If total number of free
wavelengths is y, the node selects p (y>p>1) number of
It is worth mentioning here that, control packets carry wavelengths having higher probabilities of remaining free.
information and the nodes use the information carried by the These p wavelengths are arranged with respect to probability in
control packets to execute the required tasks. But in this paper, descending order as λg1 ,λg2 , …,λgp . Here, p is a predefined
to improve the readability, we sometime describe control number which represents the maximum number of splitting
permitted for a request. Obviously, if y<=p, all y wavelengths
packets as performer of some tasks.
are selected. From the ordered set, wavelength λg1 is selected as
guessed wavelength and wavelengths λg2 to λgp are stored in
When a request arrives at source node, the node initiates the future_guess_set.
control packet PROB and forwards it to the next node towards
destination. Henceforth source and destination will be used to When the present node is the source, prev_guess_index of
represent source node and destination node respectively. PROB PROB is initialized to λg1. A record is created in node_table of
includes the following fields: connection_id, source_id, the source and PROB is forwarded to next node. If the present
destination_id, route_path, probe_map, future_guess_set and node is any node other than source, the node checks the
prev_guess_index. The route_path is the ordered list of nodes availability of the wavelength stored in prev_guess_index. If the
on the selected route, probe_map is an array indicating the wavelength is available, a record is created in node_table of the
availability of each wavelength in the route. A “1” in probe_map present node and PROB is forwarded to next node; else the node
checks for splitting. When splitting does not occur, fields: connection_id and selected_wavelength. RES_BKD
prev_guess_index is updated to the wavelength λg1. A record is includes the fields: connection_id, selected_wavelength and
created in node_table of the present node and the PROB is future_guess_set. At the point of splitting prev_guess_index
forwarded to next node. (i.e., λg1) of PROB is assigned to selected_wavelength of both
RES_FWD and RES_BKD. The RES_BKD moves towards the
2.3 Dynamic splitting source, reserving λg1 (i.e. the wavelength stored at
MMSRP adaptively splits a probe attempt into two concurrent selected_wavelength) and deleting the entries of this request in
(upstream and downstream) reservation attempts at some node_tables on the way. The RES_FWD moves towards
intermediate node selected dynamically. For a request, splitting destination reserving λg1.
may occur if both the following conditions are satisfied: (i)
PROB has traversed a pre-selected distance (x*d) of the route,
Source sp2 sp1 Destination
where d is the total distance of the route, x is a positive fraction
(0<x<1) and (ii) the wavelength at prev_guess_index is not PROB
available in present link . The effect of x on blocking probability
(bp) for various network conditions, is discussed in [8]. We RES_BKD RES_FWD
select the value of x as 0.5, considering moderate to high crisis
situation. If the conditions of splitting are satisfied, it splits;
otherwise the PROB propagates to the next node. RES_BKD

RES_FWD
3. WAVELENGTH RESERVATION IN
MMSRP
If splitting does not occur and PROB successfully reaches ACK
destination, the destination converts PROB into RES. For that
request standard DIRP is used for reserving the wavelength.
Table 1 summarizes different control packets used in the
scheme. RES includes the following fields: connection_id and Data Transmission
selected_wavelength. The content of prev_guess_index of PROB

Figure 1. Case of success in MMSRP.


Name Description
PROB moves from source to destination and contains However, if RES_BKD fails at some intermediate node due to
various probe results non-availability of λg1, further splitting may occur (maximum p-1
RES_FWD moves from splitting point towards times). When RES_BKD fails, the node selects next candidate
destination to reserve a selected_wavelength from the future_guess_set, subject to availability. The node
checks first the availability of λg2 in the forward link. If λg2 is not
RES_BKD moves from splitting point towards source to
available, λg3 is checked for its availability and so on upto λ gp. In
reserve a selected_wavelength
case of failure after first splitting, RES_BKD again splits into
ACK moves towards source, caries
two new reservation packets RES_FWD and RES_BKD. These
acknowledgement if RES_FWD reaches
RES_FWD and RES_BKD, act like earlier RES_FWD and
destination successfully
RES_BKD packets respectively. Both packets now attempt to
NACK moves towards source and caries not reserve a new wavelength selected in the same way, say λgx, in
acknowledgement if RES_BKD or PROB
both forward and backward directions. RES_FWD on its way
fails also releases the previously reserved wavelengths by previous
REL_BKD moves towards source to release the RES_FWD and RES_BKD.
wavelength reserved so far if RES_FWD fails,
also caries not acknowledgement Now, if both RES_FWD and RES_BKD are successful to
REL_FWD moves towards destination to release the reserve the same wavelength, data transmission starts after
reserved wavelength if RES_BKD fails receiving the ACK from destination. If RES_BKD is stuck at an
Table 1. Control packets. intermediate node and all possible splittings are exhausted, the
is assigned to selected_wavelength. The intermediate nodes, on request is blocked and RES_BKD is converted into NACK
receiving this RES, lock the selected_wavelength as busy and which moves towards source. Another REL_FWD is generated
delete the record from the node_table of this request. However, from the point of failure which moves towards destination and
if the selected_wavelength becomes unavailable at any point, releases the wavelengths reserved so far by both RES_FWD and
RES is converted to REL which moves towards destination and RES_BKD. Again, if RES_FWD fails, it is converted into
releases the wavelength in the links reserved so far. A NACK is REL_BKD which moves towards source releasing the
also generated from the point of failure which proceeds towards wavelengths reserved so far. It also acts as a NACK and deletes
source deleting all the records in the node_tables of this request. the entries of this request in node_tables of the path. Figure 1
If the RES reaches source successfully, transmission starts. shows a case of success whereas Figure 2 shows a case of
After the transmission is over, the reserved wavelength is failure. In the figures, sp1 and sp2 indicate the two splitting points
released. (nodes) of a request.

If splitting occurs, the PROB is converted to RES_FWD. A Source sp2 sp1 Destination
RES_BKD is also generated at the first splitting point (we call it
sp1) as shown in Figure 1 and Figure 2. RES_FWD includes the
70
PROB
RES_FWD 60 % gain over M BRP

blocking probability ( bp )
RES_BKD
50
REL_FWD
NACK 40
REL_FWD 30

20

10
% gain over M SRP
Figure 2. Case of failure in MMSRP. 0
50 100 150 200 250 300
average arrival rate of requests (cr )
4. SIMULATION RESULTS
We have tested our protocol in simulation environment and the
results are produced here. In the simulation model, static
shortest path routing is used in a randomly generated DWDM
Figure 4. Variation of % gain of MMSRP in bp with cr
network (without wavelength converters) of 40 nodes and 46
over MSRP and MBRP for wl=500.
links. Requests arrive following Poisson distribution, and
connection holding times are exponentially distributed. The
simulation model is event driven. The source and destination of
a request are chosen at random. The key performance metric bp
is studied exhaustively for different parameters. We define T ratio 0.12
M BRP
as the ratio of T ′ to T. Effect of Tratio on bp is studied in [8] and 0.1
blocking probability ( bp )
it is reported that for Tratio≈ 300, bp remains minimum. So, for
0.08
the results presented in this paper, we keep Tratio fixed at 300. MSRP
We represent the number of wavelength(s) and average arrival 0.06
rate of request(s) per second in the network as wl and cr,
respectively. Here, we have used p=2. Also, we have considered 0.04
M MSRP
MSRP and MBRP for comparison with the proposed scheme 0.02
MMSRP. We have studied the characteristics of the schemes for
higher values of wl ranging from 300 to 1000 which are suitable 0
for DWDM. However, due to limitation of space, we have 300 400 500 600 700 800 900
presented one representative result for wl= 500. We have also number of wavelengths (wl )
studied the effect of wl on bp for fixed values of cr. One such
representative result is presented here for cr=300.

4.1 Blocking probability Figure 5. Comparison of bp with wl for cr=300.


The variation of bp with respect to cr is presented in Figure 3
for wl=500. It is evident from the figure that MMSRP performs Figure 5 shows the variation of bp with different values of wl for
distinctly better than MSRP and MBRP. In general, for all the cr=300. The following points can be observed from the figure.
schemes, bp increases with increase in cr. The percentage gain First, MMSRP always performs distinctly better than MSRP
of MMSRP over MSRP is in the range of 13% to 25% and and MBRP. Second, the variation of bp with wl is very less. This
around 50% over MBRP (Figure 4). may be explained considering the two basic reasons of
0.1 connection blocking, which are: (i) insufficient network resource
M BRP and (ii) use of outdated information during the attempt of
blocking probability ( bp )

0.08 reservation. Here, with the increase of wl more resource become


available, whereby the first reason is minimized. But, due to
MSRP
0.06 propagation delay, the information used for reservation of wl
happens to be outdated. Thus the second reason exists, and
0.04 blocking takes place due to use of outdated information mainly.
M M SRP Hence, bp does not vary too much with wl.
0.02
4.2 Average setup time
0 Setup time is the latency in establishing a lightpath
50 100 150 200 250 300 corresponding to a request. Since individual setup times may
average arrival rate of requests (cr )
vary from request to request, we calculate the average of all
setup times for successful cases (Figure 6). Moreover, setup
time of MMSRP remains higher, compared to MSRP and
MBRP, throughout the region because MMSRP requires more
Figure 3. Comparison of bp with varying cr for wl = 500. time as some part of the route is traversed more than once in
case of multiple splitting in MMSRP.
0.00395 [2] C. Murthy and M. Gurusamy, WDM Optical Networks,
Concepts, Design and Algorithms. Englewood Cliffs, NJ:
Prentice-Hall, 2001.
M M SRP
0.00385 [3] D. Saha, “A Comparative study of distributed protocols for
wavelength reservation in WDM optical networks”, SPIE
setup time

Opt. Netw. Mag., vol 3, no. 1, pp. 45-52, 2002.


0.00375 M SRP [4] H. Zang, J. P. Jue and B. Mukherjee, “Review of routing
and wavelength assignment approaches for wavelength-
routed optical WDM networks”, Optical networks, Vol 1,
M BRP no 1, pp 47-60, Jan. 2000.
0.00365
50 100 150 200 250 300 [5] H Zang, J. P. Jue., L Sahashrabuddhe, R Ramamurthy, and
average arrival rate of requests (cr ) B Mukherjee,“ Dynamic Lightpath Establishment in
Wavelength-Routed WDM Networks”, IEEE Commun
Magazine, pp 100-108, Sept. 2001.
[6] K. Lu, J. P. Jue and G. Xiao, “Intermediate-Node Initiated
Figure 6. Variation of setup time with cr for wl=500. Reservation (IIR): A New Signaling Scheme for
Wavelength-Routed Networks,” IEEE JSAC, vol. 21, no. 8,
We have also studied the variation of average number of control pp 1285-1294, 2003.
packets with different values of cr for a range of values of wl [7] M. Sengupta, S. K. Mondal and D. Saha “An Adaptive Split
from 300 to 1000. Generally, the average value of control Reservation Protocol (SRP) for Dynamicaly Reserving
packets used by MMSRP is slightly higher than (or equal to) Wavelengths in WDM Optical Networks”, Proc. ICDCN
average value of control packets used by MBRP and MSRP. The 2008, LNCS 4904, pp 440-451, 2008.
increase is limited to around 7% only. This is because the [8] M. Sengupta, S. K. Mondal, C. Bose and D. Saha, “A
number of control packets used by the schemesare the same for Markov selection Split Reservation Protocol for WDM
most of the cases, such as success without split, success with Optical Networks without Wavelength Conversion”, to
single split, failure during probing, and failure during appear in Proc., IEEE GLOBECOM, 2008.
reservation without splitting. Only, for the case of more than one [9] T. Ozugur, M. Park, and J. Jue, “Label prioritization in
splitting, the number of control packets for MMSRP is more. GMPLS-centric all-optical networks,” in Proc.ICC, 2003,
Thus, the comparison of average control packets is not pp. 1283–1287.
remarkably noticeable and hence is not presented here.
[10] W. Lin, R. S. Wolff and B. Mumey, “A Markov-Based
Reservation Algorithm for Wavelength Assignment in All-
5. CONCLUSION Optical Networks”, IEEE Journal of Light wave
Technologies, vol. 25, no. 7, pp 1676-1683, 2007.
Multiple splitting, combined with multi-wavelength guessing, in

MMSRP reduces blocking probability considerably (around


15%-
50%) over either MSRP or MBRP. During probing, first
splitting is used dynamically and then multiple splitting is used
in case of failures. Hence, the average setup time in MMSRP is
slightly higher than that in MSRP or MBRP, but the average
number of control packets used is not significantly changed.
Thus, MMSRP performs better than the current best protocols as
far as blocking probability is concerned at higher wavelength
regions at the cost of nominal increase in the average setup time.
So it may be considered as a better performer in DWDM
networks, specially for the applications, where protocol
efficiency is of prime importance and the network uses a larger
number of wavelengths.

6. REFERENCES
[1] A.E. Ozdaglar and D.P.Bertsekas, “Routing and
Wavelength Assignment in Optical Networks”, IEEE/ACM
Trans. On Networking, vol. 11, no. 2, pp. 259-271, April.
2003.

S-ar putea să vă placă și