A dynamic approach to reduce preemption in book-ahead reservation in QoS-enabled networks
Introduction
In recent years, quality of service (QoS) consideration for both wireline and wireless networks has attracted renewed attention from the researchers due to growing use of high-speed multimedia and distributed applications. Recent studies show that users express great concern about QoS of the services they are paying for [10], [11]. To guarantee the QoS of applications two types of reservation techniques have been proposed by researchers: (i) book-ahead (BA) reservation (ii) instantaneous request (IR) reservation. Multimedia applications like multi party video conferencing, video on demand, live broadcast of TV programs, medical applications like remote surgery or telemedicine, tele-teaching and distance learning, etc. which require long duration, high bandwidth demand and time-sensitive significance are good candidates for BA reservations [1], [2], [3], [4], [5]. BA reservation is also required for applications like grid computing, distributed simulations that run on multiple super computers, tele-immersion applications that require simultaneous access to databases, CAD tools and rendering devices [4], [8]. Book-ahead reservation requires the resource reservation in advance and ensures that the application will have sufficient resources at the point of its start. One of the side effects of BA reservation is the preemption of ongoing IR calls to supply the required resources to a BA call if scarcity of resources arises at the starting time. Rerouting of preempted calls is often thought as a solution to maintain service continuity in a small size low speed network. However, rerouting is often costly in a large network because of the time requirement to find and establish an alternative path [26]. Also the ‘no rerouting’ restriction appears for high speed network where the bandwidth-delay product far exceeds the buffer capacity of the network [27]. This makes preemption a severe threat to service continuity which is observed as an important issue from users' perceived QoS point of view [9], [11]. This work aims to reduce preemption probability at the same time ensuring higher network utilization, lower wasted throughput and higher effective revenue.
In [3], Schelen and Pink addressed the problem of high number of IR call preemptions when resources are shared between IR and BA reservations. They introduced the concept of look-ahead time to reduce the problem of high preemption probability of IR calls. Look-ahead time is defined as the pre-allocation time, i.e. the time for starting to set aside resources for BA reservation so that there is no resource scarcity at the starting time of a BA call. Schelen et al. considered constant look-ahead time (CLAT) and presented simulation studies based on IR preemption probability and resource utilization at different constant values of look-ahead time. With CLAT scheme preemption probability decreases with larger look-ahead time at the cost of lower network utilization. Keeping the look-ahead time constant is not logical in a dynamically changing environment. A systematic analytical method to determine the look-ahead time taking the traffic conditions in consideration is not dealt in Ref. [3]. Wischik and Greenberg proposed an admission control scheme for BA calls which estimates effective bandwidth and thus provides only statistical guarantee [13]. Their proposed model is based on preemption probability of BA calls where the BA call with the shortest BA time is preempted in case of congestion. Statistical guarantee is a good idea from a network's point of view, but deterministic guarantee provides certainty which is easy to understand from a user's perspective and hence popular among the users. In Ref. [15], Degermark et al. assumed that all calls, whether booked in advance or immediate, declare their duration in advance. Quantitatively IR calls are expected to dominate a commercial network and hence it is unreasonable to require that the duration of each IR call be specified in advance which is often context dependent and unforeseeable.
Some researchers have proposed strict partitioning of link capacity that divides the network resources into two disjoint subsets dedicated to each class of calls and thereby eliminates the problem of preemption of any on-going calls [12], [14]. But optimizing the partition usages is difficult and may drastically reduce the network utilization. Greenberg et al. [2] proposed an admission control scheme based on an approximate calculation of interrupt probability, which shows that resource sharing even at the cost of some preemption gives better utilization than that in strict partitioning of resources. The scheme proposed in Ref. [2] depends on correct individual prediction of future incidents like call service rate and holding time. As mentioned by Schelen et al. [3] and also supported in our observation, prediction of holding time of IR calls and accordingly using look-ahead time based on predicted holding time performs inferiorly as compared to CLAT based model. In [1], Lin et al. proposed an application aware look-ahead time based admission control scheme which attempts to predict call holding time for each IR call based on which application category it belongs and takes the predicted holding time as the look-ahead time using a traffic pattern of the link. This scheme is entirely dependent on the traffic pattern of individual link, which is often difficult to gather and may vary widely from link to link. In practice, a call connection traverses hundreds of links with different traffic patterns from source to destination, which makes the application sensitive scheme practically infeasible. Prediction of future incidents like call holding time (both exact and probabilistic) has been considered as a hard issue by researchers and this is why most of the works assume call holding time of an IR call as open ended and make their approach not sensitive to call holding time [5], [6], [7]. These considerations make the model proposed in [3] more practical and generalized for BA reservation in a QoS-enabled network. However, this model employs constant look-ahead time for all IR calls which does not stand valid in real time networks. No model has been proposed so far to employ an adaptive look-ahead time that reflects the network dynamic nature. This paper presents a novel technique to calculate look-ahead time incorporating the dynamic nature of a network. Our work aims at achieving low preemption probability at the cost of marginal loss of throughput and wasted revenue, but yields higher effective throughput and revenue at the comparable level of preemption probability in CLAT scheme. Simulation result shows that the proposed model outperforms CLAT-based model from QoS perspective.
The paper is organized as follows. In Section 2, we present the problem scenario and existing model to solve the problem. In Section 3, we introduce the proposed model. Simulation results are shown in Section 4. Section 5 deals with the parameter selection related to the model followed by concluding remarks in Section 6.
Section snippets
Resource reservation and preemption of IR calls
A BA call needs to announce its starting time and call holding time along with QoS parameters. Call admission control (CAC) algorithm at each node along the path from source to destination checks whether there will be enough resources for that BA call at the announced starting time and for the announced duration [16]. If CAC finds it feasible, it allocates the resources for the whole announced duration. A book-ahead table is maintained to ensure that over-allocation of resources is never done.
Dynamic look-ahead time (DLAT) based model
The objective of the proposed model is to calculate the look-ahead time dynamically with the network condition. Both IR and BA load vary with time. At any operating time in a network the exact BA load in near future is known from the book-ahead table, while IR load can be approximated from the current trend of IR load. A format of a book-ahead table is shown in Table 1. It contains the actual start time of BA applications, the aggregate bandwidth A(s) that is reserved for BA calls and
Simulation results
In order to evaluate the performance of the proposed DLAT model, we conducted simulation with different network parameters and made a comparison with CLAT model. As stated earlier the focus of our experiment was to investigate how the model performs in terms of preemption probability, network utilization, wasted throughput and call blocking probability.
Estimation of c w.r.t. preemption probability in DLAT model
As discussed in Section 3, traffic parameters may subject to change at different operating periods. For a dynamic system like communication network, periodic data analysis provides relatively better picture of concurrent network scenario. Dynamic look-ahead time and hence preemption probability in the proposed method is influenced by changing traffic parameters (Eq. (8)). Desired level of preemption probability can also vary depending on the traffic type. From the operational point of view it
Conclusion
Book-ahead (BA) reservation is likely to play an important role in a QoS-enabled network for better resource management. This paper addresses one of the major problems for successful implementation of BA reservation in real-time QoS-enabled networks. When resource scarcity arises, BA calls cause preemption of on-going IR calls and higher preemption rate of IR calls compromises the perceived meaning of QoS and users' satisfaction. Although rerouting of preempted calls is often thought as a
References (29)
- et al.
A proposal of new price-based call admission control rules for guaranteed performance services multiplexed with best effort traffic
Comput. Commun. Trans.
(2003) - et al.
Comparative analysis of path computation techniques for MPLS traffic engineering
Comput. Netw.
(2002) - et al.
Bandwidth brokers of instantaneous and book-ahead requests for differentiated services networks
ICICE Trans. Commun.
(2002) - et al.
Resource sharing for book-ahead and instantaneous-request calls
IEEE/ACM Trans. Netw.
(1999) - et al.
Sharing resources in advance reservation agents
J. High Speed Netw.
(1998) - et al.
Networks with advance reservations: the routing perspective
(2000) - et al.
Concepts for resource reservation in advance
J. Multimedia Tools Appl.
(1997) - et al.
DRES: network resource management using deferred reservations
Proc. GLOBECOM'01
(2001) - et al.
Issues of reserving resources in advance
Proc. NOSSDAV
(1995) - L. Burchard, Source routing algorithms for networks with advance reservations, Technical Report, TU Berlin, ISSN...
QoS Measurement and Evaluation of Telecommunications Quality of Service
Engineering Internet QoS
Multiservice Loss Models for Broadband Telecommunication Networks
Cited by (11)
Rerouting in advance for preempted IR calls in QoS-enabled networks
2008, Computer CommunicationsOn Hybrid IR and AR Service Provisioning in Elastic Optical Networks
2015, Journal of Lightwave TechnologyAnalytical blocking probability model for hybrid immediate and advance reservations in optical WDM networks
2013, IEEE/ACM Transactions on NetworkingImproving service differentiation of immediate and advance reservation in resource-partitioned optical WDM networks
2013, 2013 International Conference on Computing, Networking and Communications, ICNC 2013Application of artificial intelligence to improve quality of service in computer networks
2012, Neural Computing and ApplicationsAdvance and immediate request admission: A preemptable service definition for bandwidth brokers
2012, International Journal of Computers, Communications and Control