Bankruptcy-based Radio Resource Management for Multimedia Mobile Networks

This is the pre-peer reviewed version of the following article: Lucas-Estañ MC, Gozalvez J, Sanchez-Soriano J. Bankruptcy-based Radio Resource Managem...
Author: Camron Bradley
2 downloads 0 Views 346KB Size
This is the pre-peer reviewed version of the following article: Lucas-Estañ MC, Gozalvez J, Sanchez-Soriano J. Bankruptcy-based Radio Resource Management for Multimedia Mobile Networks. European Transactions on Telecommunications, Volume 23, Issue 2, pages 186–201, March 2012| DOI: 10.1002/ett.1525 which has been published in final form at

Bankruptcy-based Radio Resource Management for Multimedia Mobile Networks M.C. Lucas-Estañ#1, J. Gozalvez#2, J. Sanchez-Soriano*3 #

Ubiquitous Wireless Communications Research (Uwicore) Laboratory, University Miguel Hernandez of Elche, Spain * Operations Research Center, University Miguel Hernandez of Elche, Spain, 1

[email protected], [email protected], [email protected]

Abstract—The transmission of bandwidth demanding multimedia applications in capacity constrained mobile radio networks requires optimizing the usage and assignment of radio resources following the varying quality of service requirements characteristics of multimedia traffic environments. Considering the capacity of bankruptcy theory to deal with situations where the demand for resources is higher than its availability, this work proposes the application of bankruptcy theories to design efficient radio resource management policies that provide the highest possible Quality of Service levels and guarantee user fairness. Index Terms— Bankruptcy theory, mobile radio networks, multimedia traffic, Radio Resource Management. I. INTRODUCTION The increasing introduction and demand for multimedia services in mobile and wireless systems is urging mobile operators to develop adequate Radio Resource Management (RRM) policies to efficiently use the scarce available radio resources. The simultaneous allocation of multiple radio resources to a given user can increase its throughput and reduce its end-to-end delay. However, considering the limited number of radio resources, the

performance improvements could be produced at the cost of blocking new users, or unfairly assigning radio resources among users. In this context, it is important to design assignment policies that avoid channel stagnation while considering the system load and the user Quality of Service (QoS) requirements. On the other hand, different traffic services might have varying QoS requirements. These varying QoS and resource demands could be exploited to design efficient and fair radio resource distribution mechanisms in multimedia traffic scenarios. To design adequate radio resource distribution mechanisms, several factors should be taken into account. First of all, their computational cost should be feasible to allow for their real implementation. The work reported in [1] showed that assignment schemes that consider all possible distribution patterns result in excessive computational costs that currently prevent their use in mobile communication systems. Radio resource distribution policies should also be able to assign the adequate number of resources to each user based on their QoS demand and the system load. To this end, the work reported in [2] highlighted the performance improvement in terms of admission probability and dropping probabilities achieved by Admission Control, Packet Scheduling and Congestion Control mechanisms based on QoS differentiation. The consideration of user fairness is also becoming an

important objective of current RRM techniques. An interesting study addressing the concept of user fairness is found in [3]. This work analyzed the trade-off between system spectral efficiency and fairness among users that could be achieved when different RRM techniques employed in multi-carrier cellular systems are used. In this work, the user fairness is measured as a function of the throughput achieved and required by each user. The evaluation was carried out in OFDMA-based networks with only non-real time services. A scheduling algorithm for future wireless systems designed to balance between the requirements of the different users and the requirements of service providers is also reported in [4]. The authors of [4] proposed utility functions that try to ensure a fair distribution of the radio resources among users, where the user fairness is defined in terms of the distribution of average throughput per user. While these papers aimed to achieve equal radio resource or throughput assignments, the RRM techniques proposed in this work have been designed to achieve homogenous QoS levels to users in multimedia environments, where different traffic or service classes might have varying QoS requirements. Other approaches considering the use of utility-based radio resource distribution policies were suggested in [5] and [6]. The radio resource distribution proposed in [5] is based on utility functions that describe the relationship between the user’s service appreciation and the allocated resources, and also on the price to pay for the resources. This interesting study is focused on maximizing the provider’s revenue, and also introduces the use of backward utility functions, which represent the effect of service degradation that may happen when a new connection is admitted in dynamic resource assignments. On the other hand, [6] uses utility functions based on the price users are willing to pay for radio resources, and proposed the use of economic concepts to address the radio resource assignment problem by defining Vickrey auctions derived to satisfy QoS requirements to users only considering constant bit rate services. Users participate in the auctions with their utility functions, and an auction is made for each resource. This mechanism resulted in that users achieved very high satisfaction levels, or were not assigned any resources. Of particular interest is also the proposal reported in [7]. This work proposed a radio resource assignment scheme that seeks to fairly distribute available resources. To establish their distribution policies, the authors first estimate the bandwidth needed by a user to guarantee an adequate transmission. Based on such estimates, the reported scheme equally distributes the available

bandwidth among users. If all users are satisfied, and there is yet available bandwidth, the available resources are again equally distributed among users. However, it is important to note that this proposal distributes frequency bands, which is a resource considered continuous and infinitively divisible. On the other hand, this work considers the assignment problem based on radio resources (timeslots or codes) of discrete nature, and the resource distribution policy is aimed to fairly achieve homogeneous QoS satisfaction levels for all users in the system. In this context, this paper proposes to further expand the use of economics concepts to address the radio resource assignment problem in multimedia traffic scenarios by developing innovative policies using bankruptcy theory. In bankruptcy situations, the value of a company is inferior to the sum of its debts, and adequate distribution policies are needed to divide its net value among its creditors [8]. Given the similarity of the problems dealt with by bankruptcy and the radio resource assignment distribution dilemma in highly loaded multimedia traffic environments, this work extends the work carried out in [9] and [10], and presents bankruptcy-based radio resource assignment proposals designed to maximise the radio resource efficiency and user satisfaction fairness, while exploiting the varying QoS demands from different service classes in multimedia networks. To this aim, the bankruptcy distribution rules must be adapted to work with discrete resources. In addition to considering resources of discrete nature, this work also considers the use of utility functions that are not directly proportional to the number of resources allocated to a user. To provide a framework over which to compare the performance and operation of the bankruptcy-based proposals, this work also implements the Maximum Utility Increase (MAUI) and Required dATa ratE (RATE) policy schemes. Since some of these proposals base their decision criteria on traffic service-dependent utility functions, the definition of these utility functions is first presented. II. RADIO RESOURCE ASSIGNMENT PROPOSALS A. Traffic Service Utility Functions Utility functions try to characterize the QoS satisfaction level experienced by a user according to the number of assigned radio resources and the requested traffic service. This is a challenging task because user satisfaction is a subjective concept that heavily depends on user perceptions. This work considers a multimedia traffic scenario with email (background), WWW

(interactive) and real-time H.263 video (at 16, 32 and 64kbps mean bit rates) users. Following 3GPP recommendations [11], this work considers email and WWW transmissions satisfactory if an email or web page is transmitted in less than 4 seconds. For real-time video services, video frames are considered to be satisfactorily transmitted if they are completely transmitted before the next video frame is to be transmitted. To define the traffic utility functions, the user QoS levels reported in Table I have been chosen1. These minimum, mean and maximum QoS levels are based on user throughput for web and email transmissions, and on the percentage of correctly transmitted frames for real-time video services (the video utility function has been chosen to be independent of the H.263 video bit rates). For real-time video services, the maximum utility value can be achieved when all video frames are transmitted before the next video frame is to be transmitted. On the other hand, the retransmission capability of non-real time services results in that even high transmission rates might not be able to satisfactorily transmit a web page or email following the 3GPP recommendation (i.e., in less than four seconds). In fact, the transmission time for non-real time services not only depends on the data rate, but also on the object size and the channel quality conditions. As a result, this work avoids the definition of a utility value equal to one for non-real time services requiring high transmission reliability levels (e.g., email and web). It is also important to note the high requirements (percentage of video frames) established for real-time H.263 video services. Following the H.263 video model proposed in [12], I-frames coding a complete video image must be regularly sent to provide a good user QoS perception. Given their important size, and the mobile bandwidth limitations, I-frames could experience a high probability of not being completely transmitted before the next video frame is to be transmitted. To avoid such scenario, high QoS levels have been selected for the H.263 video utility functions. This fact has also motivated the different shape of the utility functions for real-time and non-real time users (Fig.1). For non real-time services, this work has defined a utility function that linearly grows between the minimum and maximum QoS levels. Once the maximum QoS level is reached, the utility value slowly increases with the throughput to account for the possibility of not satisfactorily transmitting a web page or email in less than 4 seconds due to bad channel 1

This is an arbitrary choice. Different values could have been chosen without modifying the operation and performance trends of the proposed techniques.

TABLE I USER QOS LEVELS Min. QoS Mean QoS Email 16kbps 32kbps WWW 32kbps 64kbps Established utility values 0.95/4 0.95/2 H.263 video 75% 95% Established utility values 0.95/4 0.95/2

Max. QoS 64kbps 128kbps 0.95 100% 1

conditions and the retransmission capability of non-real time services. On the other hand, the utility function for real-time services slowly increases (between the minimum and mean QoS levels) until a high percentage of transmitted video frames is guaranteed. This behaviour has been chosen given the indications in [13] that highlight that an acceptable video quality requires a high percentage of correctly received video frames. It is important to note that for all services, utility values equal to zero are assigned if the minimum QoS requirements cannot be achieved. This avoids assigning resources to users that would experience very poor QoS levels. As previously mentioned, we would like to highlight that the definition of these utility functions is a challenging task because user satisfaction is a subjective concept that heavily depends on the user’s perception. The utility functions used in this work try to follow certain indications reported in the literature, e.g. those related to video quality and user perception [13] and 3GPP indications [11]. The definition of optimal utility functions relates more to the research domain of QoS human perception, which is beyond the authors capabilities and out of the scope of this research. Although the use of different utility functions would modify the absolute performance values, it would not influence the operation of the bankruptcy-based proposals, and the performance trends reported in this work. The work reported in this paper considers the GERAN (GSM/EDGE Radio Access Network) radio interface as a TDMA representative technology over which to test the paper’ proposals. Of course, the proposed bankruptcy-based techniques can be used and adapted to other standards and multiple access technologies requiring the definition of policies to distribute multiple radio resources among active users. However, this work is focused on the demonstration of the benefits that can be achieved through the use of bankruptcy-based techniques in radio resource management problems, and therefore the adaptation to other particular systems, although possible, is out of the scope of the paper. In GERAN, a radio resource corresponds to a TDMA timeslot (TS). GERAN allows the allocation of multiple

si) is higher than one for certain traffic services. This is due to the established QoS levels reported in Table I, and the fact that the implemented policies try at least to achieve the minimum QoS level for each user. As a

result, each service class requires a different number of resources to achieve a utility value greater than zero. Furthermore, the discrete utility functions satisfy that if 0 < Us(r) < 1, then Us(r) < Us(r+1). This property means that when a user begins to perceive positive satisfaction, then every additional resource leads to a strictly positive satisfactory increment, except when its maximum QoS level (maximum utility value) is achieved. It is important to note that although different utility values would be needed if modifying the system operating conditions, the multimedia traffic scenario or the employed radio interface, the proposed methodology to establish the utility functions would still be valid. In addition, the comparison of the bankruptcy based techniques with the RATE (Required dATa rate) proposal (defined in Section II.D) will highlight the adequate definition of the utility functions based on user needs. B. Bankruptcy-based Radio Resource Assignment Bankruptcy theory has been proposed to solve situations where the number of claimed resources exceeds those available. In these cases, policies are defined to decide how scarce resources should be fairly distributed among the claiming creditors. Due to the high similarity with the radio resource assignment dilemma under high user load scenarios, this paper evaluates the potential of the bankruptcy theory to define efficient and fair radio resource assignment policies. In 1


0.6 0.4 0.2 0

real-time video

0.8 Utility value

Utility value


100 200 Throughput (kbps)

0.4 0.2

WWW service email service 0



0 60 70 80 90 100 Correctly transmitted video frames (%)

Fig.1. Utility functions per traffic service.

1 TS

2 TS

3 TS

4 TS

5 TS

6 TS

7 TS

8 TS





radio resources to a single user, with the maximum number of resources that can be assigned being equal to eight timeslots. GERAN also implements an adaptive radio interface that dynamically varies the used transmission mode (modulation and coding scheme, MCS) according to the experienced channel quality conditions. Such adaptive operation and the continuously varying channel quality conditions make difficult to predict the mean bit rate that can be achieved with a given number of radio resources. In order to be able to relate the utility functions to the number of assigned radio resources, this work considers the MCS5 transmission mode, which provides a mean bit rate of 22.4kbps per timeslot. This transmission mode is selected since it has been identified as the most widely used transmission mode in previously conducted system level simulations conducted considering the system scenario reported in Section III. However, it is important to note that different system configurations would require identifying again the mostly used transmission mode, and consequently a redefinition of the considered utility values, although this would not influence the obtained performance trends. With the selection of a transmission rate per timeslot, the relation between assigned radio resources and utility values can be directly obtained through Fig. 1 for email and web services. Such relation is reported in Fig. 2. For real-time H.263 video services, an additional step is necessary. A cumulative distribution function (CDF) of the throughput needed to transmit each video frame before the next video frame is to be transmitted is derived following the H.263 video model proposed in [12]. Through these CDFs, the percentage of video frames reported in Fig. 1 can be related to the corresponding necessary throughputs for the various video bit rates considered in this work (16, 32 and 64kbps), and consequently to the necessary radio resources considering the MCS5 transmission mode to estimate the throughput achievable per radio resource. The final discrete utility values (Us: N  [0, 1], where N is the set of all non negative integer numbers) defined per traffic service (s) and number of assigned radio resources to a user (r  N) are shown in Fig. 2. These utility functions satisfy that Us(0)=0 and Us(r) ≤ Us(r+1). It is important to note that the minimum number of radio resources needed to achieve a positive utility value (li = min{r  N: Usi (r) >0} for user i demanding service






16kbps video

32kbps video

64kbps video

Fig.2. Utility values per traffic service and number of radio resources.

particular, this paper evaluates the potential of bankruptcy theory through the adaptation of two bankruptcy rules, Constraint Equal Award (CEA) and Constraint Equal Loss (CEL) [8], which were originally proposed to achieve fair distributions of scarce resources. For the formal definition of a bankruptcy situation, let’s consider a value E  R+ that has to be divided among N agents with claims adding up to more than E. For each i  N, ci  R+ represents the resource’s request by i, and c ≡ (ci)i the requests vector. In this context, a resources distribution dilemma is denoted by (c, E)  R+xR+ such that Σci ≥ E. Traditionally, bankruptcy theory has been applied to the distribution of continuous resources. However, the definition of bankruptcy-based radio resource assignment policies for mobile radio networks needs to work with radio resources of discrete nature. It is important to note that, considering the development of bankruptcy theory over the years, only recently has been addressed the extension of bankruptcy distribution rules to handle discrete resources. An example is the work reported in [15], although it is based on utility values that are directly proportional to the number of assigned resources (as in the original bankruptcy rules). On the other hand, the resource distribution techniques proposed in this work are designed considering utility values that are not directly proportional to the number of assigned radio resources (Section II.A). To the author’s best knowledge, this is the first work that considers the extension of bankruptcy policies to distribute discrete resources considering a variety of user profiles and requests; in addition, it is the first contribution considering the application of bankruptcy-based techniques to address radio resource management problems. This working assumption provides a better representation of multimedia wireless scenarios, where users from different service classes might experience unequal QoS satisfaction levels even if receiving the same number of radio resources. 1) General Framework In mobile radio networks with a finite number of available radio resources, R, a finite set of users, N, claim for different services (s  S, where S is the set of all traffic services offered in the system). Each traffic service can be characterized by varying QoS requirements and resource demands and, in general, the QoS perceived by users requesting different services is not directly proportional to the number of assigned resources. This work proposes the use of the utility functions defined in Section II.A (Us: N [0, 1]) to

quantify the radio resource requirement of each service class to achieve different satisfaction levels. One of the sought objectives of the proposed bankruptcy-based policies is user fairness. Achieving this objective can be challenging due to the discrete nature of radio resources and the potentially high number of users in a system. In this case, the system establishes a user priority criterion represented by

Suggest Documents