three-dimensional TV (3DTV ), digital cinema, interactive gaming, e-health, e-science, cloud computing, banking data backup storage, and grid computing to ...
AF4A58.pdf
ACP Technical Digest © 2012 OSA
Improving Connections Provisioning in Hybrid Immediate/Advance Reservation WDM Networks Ajmal Muhammad, Robert Forchheimer Linkoping University, Linkoping, Sweden, Email: (ajmal.robert}@isy.liu.se
Abstract:
We study a dynamic WDM network supporting applications with immediate
and advance reservation (IRlAR) requirements. To diminish the adverse effect of AR on IR connection provisioning, we propose scheduling strategies able to significantly reduce IR blocking probability. © 2012 Optical Society of America OCIS codes: (060.1155 ) All-optical networks; (060.4256) Networks, network optimization; (060.4265) Networks, wavelength routing 1.
Introduction
The emergence of new applications such as video on demand (VoD), distribution of ultra-high definition TV (UHDTV ), three-dimensional TV (3DTV ), digital cinema, interactive gaming, e-health, e-science, cloud computing, banking data backup storage, and grid computing to mention a few, are pushing the
wavelength division multiplexing (WDM)
to expand from a network core technology towards the access networks [1]. It is envisioned that optical networks equipped with agile devices, such as
reconfigurable optical add-drop multiplexer (ROADM) and tunable transceivers
integrated with G-MPLS/ASON control-plane technology will be extended to every premises. This will provide a platform for the above mentioned applications that hold the salient features of being user-controlled, bandwidth intensive, and with relatively short connections but with known duration. These applications can be classified into two types, namely, delay-sensitive and delay-tolerant applications. Delay-sensitive applications require immediate provisioning and allocation of network resources, for which numerous dynamic on-demand connections provisioning schemes have been developed. However, delay-tolerant applications (e.g., banking backup data, grid computing, and e-science) require network resources availability before a set deadline, thus an advance reservation (AR) technique can be very beneficial for enhancing their provisioning. In AR the resources are reserved prior to when they are actually utilized. The time interval between reservation and utilization of the network resources is denoted as the
book-ahead
(BA) time. The BA time enables AR connection requests to find network resources for their provisioning, as the network is mostly free in the far ahead future time. The AR connection requests will coexist with the on-demand, i.e.,
immediate reservation (IR) connection requests and both traffic demands will share the network resources. In
the coexistence scenario, the temporal advantage of AR connection request due to BA time exacerbates the blocking probability (BP) of IR connection requests. This is because the network resources are already occupied by the AR requests when a new IR request arrives to the network. This imposes the need to investigate how to promote the provisioning rate of IR connection requests in the mixed traffic scenario. Furthermore, the AR traffic inflicts resource scarcity for IR requests by a factor that depends on the AR traffic portion within the total network traffic. Conversely, the provisioning of IR connection requests have no significant effect on AR requests provisioning if the BA time is far greater than the connections holding-time. This necessitates the need to examine whether shared resources scenario is only beneficial for the AR class, or the IR traffic can also gain from shared resources under certain conditions. The objective of this work is to explore some new strategies for enhancing the provisioning performance of IR connections requests without compromising the improvement in terms of AR traffic blocking performance and network resource utilization. The remainder of this paper is organized as follow. Section II describes the problem statement and notations. The proposed strategies for promoting the IR traffic provisioning are explained in section III. Simulation results are shown in section IV, and finally some concluding remarks are made in section V. 2.
Problem Statement and Notations
We consider a centralized controlled WDM network with IR and AR connection requests, each request requiring one wavelength of capacity. The dynamic provisioning of WDM connections with different temporal scheduling require ment can be formulated as follows:
Given: a) Physical topology of a network represented by a graph G with a set of links E and nodes V; maximum e denoted by W; b) a connection request R {s,d,ta, a,th} between a source destination pair {s, d} with arrival time ta, book-ahead time a I, and holding-time tho Find: A route and wavelength for the connection request for the time window [ta + a,ta + a + th]' number of wavelengths on each link
1 IR
can be treated as a special case of AR with zero book-ahead time, i.e.,
=
a =
O.
AF4A58.pdf
ACP Technical Digest © 2012 OSA
Objective: Promote the acceptance rate of IR connection requests along with improving the network resource utiliza tion. 3.
Proposed Strategies for Improvement of Immediate Reservation Connections Provisioning
This section describes a set of strategies for IRiAR connections provisioning with the objective to promote the con
maximum current utilization [2]. The algorithm exploits the connection th knowledge to compute k-shortest
nection requests of the IR class. For the RWA problem this work uses an algorithm called (MCU) for its better BP performance
paths between the requested source destination and selects a free wavelength that has high network utilization within the new request 3.1.
th window, and consequently the path that hold the selected wavelength.
MCU with
Different Paths set (DP)
This strategy utilizes different values of k for IR and AR connection request while making use of the MCU for RWA calculation. To promote IR connection requests, the value of k is adjusted to a larger value than that for AR connection requests. Larger value of k enhances the search space and thus improves the probability for IR requests to find free wavelength and get connected. 3.2.
MCU with
Rerouting
This strategy employs the rerouting technique once the MCU fails to find free wavelength for an IR connection request. The rerouting technique adopted here computes the set of connections L, whose paths overlap with the k-shortest paths
[ta,ta + th]' The computed connections may be either already active [ta,ta + th]. The network resources occupied by connections L are
of the IR connection request for the time window or will become active within the time window
released (only conducted in the algorithm and does not imply the release of physical connection), and the availability of resources for IR connection request are re-checked. If resources for IR connection request become available, then new RWA solutions for connections L are calculated using MCU. However, if new RWA does not exist for any of the L connections, then they are restored on their original wavelengths and the IR connection request is blocked.
Lightpath Switching (LPS) [3] mechanism allocates time-slots on different wavelengths and paths to a given con nection request, to get connected if no wavelength is free for the entire duration of the connection tho To improve the
3.3.
MCU with
The lightpath switching (LPS)
connection provisioning for IR class requests, this strategy incorporates the LPS with MCU, i.e., if the MCU fails to find a free wavelength for the given th window of an IR request on all k-shortest paths, then the LPS strategy is utilized. The LPS technique employed here switches only the wavelength while keeping the same path during the connection tho The LPS starts scanning with the first shortest path and lowest-index wavelength for free time-slots which are turned into segments. O nce the current wavelength is scanned for the
th window, i.e., [ta,ta + th], it moves to the next higher
index wavelength and searches for free time-slots that do not overlap in time with the previously selected time-slots. The scanning process terminates once it finds the free time-slots for the entire th window of an IR connection request. To evaluate the impact of routing, the LPS technique is further classified into two schemes, namely, LPS dynamic routing (LPS-D) and LPS static routing (LPS-S). The LPS-D makes use of the k-shortest paths computed by the MCU, while LPS-S employs the pre-computed (i.e., offline) k-shortest paths.
Delay Tolerance (DT) [4] specifies the time td up to which a connection can be delayed or kept on waiting until it is set-up. The waiting time td is expected to be a small fraction of connection request th, and the connection request
3.4.
MCU with Set-up
Set-up delay tolerance (DT)
is provisioned immediately after the required resources become free (due to the departure of an already provisioned connection) within its waiting time. To enhance the accommodation ofIR connection requests, this strategy utilizes the DT technique if the IR connection requests can not be set-up at their
ta due to wavelength unavailability. Furthermore,
like LPS the DT technique is also considered for dynamic (DT-D) and static (DT-S) routing schemes. 4.
Numerical Results
For performance evaluation of the proposed strategies we custom-built a discrete event-driven simulator and simulated the proposed strategies for the European optical network with 30 nodes and 48 bidirectional links is divided into fixed time-slots of width �, and
(x,
[5]. The time-domain
connection th, and the set-up delay tolerance are integer multiples of
�. The network links are assumed to be bidirectional with the same number of wavelengths, W=16, in both direction.
It is further assumed that the network is not equipped with wavelength convertors, thus a connection request must use the same wavelength from the source to the destination node. Connections are uniformly distributed between AR and IR, and are assumed to arrive in the network according to a Poisson process with connection
th following
an exponential distribution with mean equal to 10 units. The source and the destination of a connection request are uniformly distributed over the set of network nodes, and the value of k is set to 3 except for the IR class in DP strategy for which its value is fixed to 6. For DT strategy, the value of
td is set to 10% of connection holding-time. Uniform
AF4A58.pdf
random distribution is used to generate
ex
ACP Technical Digest © 2012 OSA
within the interval [25,45] for AR connections. To investigate the impact of
AR class traffic over theBP performance ofIR class traffic, theBP ofIR class in mixed traffic scenario is compared for the MCU algorithm, in the Fig. 1(a), with IR performance when there is no AR traffic. However, the network resources for the latter case are reduced by 50% (to account for absence of AR traffic that constitutes 50% in the mixed case), i.e.,
8 wavelengths per link. Note that the traffic in Fig.l(a) presents the IR class traffic within the network, while the
total network traffic for the mixed scenario will be twice that of IR traffic. The figure depicts that at low loads the BP performance ofIR class in mixed scenario is better than its performance when no AR traffic exist. This reveals that the IR traffic can benefit from the shared resources at low loads. However, as the network load increases the performance of IR class in the mixed scenario starts to deteriorate, which indicates that AR traffic benefits from shared resources at medium and high loads. The proposed strategies are compared in Fig.l (b) and Fig.l (c) with the basic approach which is MCU for benchmarking purpose. Fig. 1(b) displays the total network BP, whereas Fig.l (c) exhibits the normalized value of IR class BP for three different traffic load cases. In Fig.l(c), the value of BP for the proposed strategies is normalized to the value of BP of MCU for all load levels. According to the performance results shown in Fig.l (b) and Fig.l (c), all the strategies achieve the stated goal, which is to reduce BP for IR class without compromising the overall network performance. The improvement in the overall network performance occurs as a result of the IR class, as these strategies have no significant effect on the BP of AR class, which are not shown here due to space limitation. Among these strategies, DP, which augments the search space for IR class traffic to find free wavelengths, demonstrates the best performance. Furthermore, it can be seen that the performance of LPS and DT improves while employing the static routing instead of dynamic routing. This is because, dynamic routing incorporates the network active connections while computing the routes and thus the routes may be longer than the static routing. Therefore, the chance for acquiring network resources on shorter (static) routes is higher compared to dynamic routes, when the LPS and DT strategies are utilized once the MCU fails to find a free wavelength.
'1
12,-----------,-��-� ___, -e- Onl)' IR traffic scenario
o
,:", 10
�H 1"
'
-¢- ' Mixed traffic
,;' ��
:i l.5 �
e; 0.8 .5 tl 0.6 � 0.
.LPS-D � 2S . .LPS_S e 2_DT-D �, .DT-S
6
�
.5 '13 _MelJ -;:: 31 IRerouting ::=
scenario
4
� :c
� :c
2
�
�=!���������� � � � � NO � 60 � NO
� 0.5 Offered traffic (Erlangs)
(a) IR BP
vs.
IR load
;-
"C
��
1
0
110
I
•••••
130 Offered trame (Erlangs)
(b) Total BP
vs.
5
1 0
network load
=
:::
1
4
0.2
130 Offered traffic (Erlangs)
(c) IR BP vs. network load
Figure 1. BP vs. load 5.
Conclusions
IR and AR reservation techniques can be employed to provision delay-sensitive and delay-tolerant applications, re spectively. However, in shared network resources at medium and high loads the book-ahead time enables the AR connections to occupy most of the resources resulting in aggravating theBP ofIR connection. To mitigate this adverse effect, some strategies for promoting IR connection requests are presented. Among the presented strategies, the one that utilizes larger path set to search free resources for IR connection request (i.e., MCU with DP) is the most promis ing strategy. Moreover, the MCU technique exhibits best performance for holding-time aware dynamic routing [2]. However, once the LPS and DT strategies are employed when an IR connection request fails to seek free resources, then in the static routing performs better than the dynamic routing scheme. Finally, these strategies will enable the network operator to select a suitable scheme based on the users demand, pricing policy for IR traffic, and operational overhead.
References 1. L. Gutierrez et aI., "Next generation optical access networks from TDM to WDM," Trends in Telecommunica tion Technology, (2010). 2. A. Muhammad et aI., "Coexistance of advance and immediate reservation in WDM networks-some RWA strate gies," ICTON, (2012).
3. N. Charbonneau et aI., "Dynamic circuits with lightpath switching over wavelength routed networks," IEEE ANTS, (2010). 4. C. Cavdar et aI., "Shared-path protection with delay tolerance (SDT) in optical WDM mesh networks," IEEE/OSA JOLT, (2010). 5. R. Inkret et aI., "Advance infrastructure for photonics networks: Extended final report of COST Action 266".
