Resource scheduling strategy for uncertain factors in mobile cloud computing

Abstract

At present, resource configuration of mobile cloud computing has received extensive attention from the outside world. Most of the similar resource scheduling configuration fails to comprehensively consider the dynamics of mobile terminals and the difference in user requested resources. Therefore, considering uncertainty in paging scheduling under mobile cloud resource environment from the perspective of consumers has become the key to solving the problem of resource allocation in the mobile cloud computing environment. This paper proposes an adaptive matching resource allocation algorithm based on uncertain factors under mobile cloud computing environment. Uncertain factors of the mobile terminal are derived via QoS attribute, and then user information and load characteristics of the user requested resources are analyzed through CLIQUE similarity matching. Afterwards, based on the mapping between similarity and resources, resource paging allocation can be carried out based on adaptive matching resource allocation algorithm. From the perspective of consumers, dynamics of mobile terminals and uncertainty of paging scheduling in the mobile cloud resource environment under different user requested resources can be considered to allow minimized delay and optimized paging strategies.

Keywords

Mobile cloud computing uncertain factors resource scheduling paging strategy

1 Introduction

The uncertain factor that produces mobile cloud computing is mainly that: with the rapid development of mobile cloud communication networks over the years, mobile infrastructure is gradually improved, but still needs to be strengthened; the uncertainty of user requested resources is mainly manifested in the difference in the gathered information under the same network, specifically due to uncertainty of network speed; the uncertainty of paging strategy;

The uncertainty of QoS with regard to mobile terminal performance is specifically manifested in the following aspects: 1) The impact of QoS attributes on mobile terminal performance is not considered; 2) The relationship between weights and node attributes is not considered at matching; 3) In resource allocation, the number of virtual machines used and the use of minimum free space of virtual machines are not considered.

With the rapid growth of mobile communication demand, mobile communication technology continues to develop and progress in the trend of technological development. While guaranteeing good data volume analysis and business volume adaptability, mobile cloud communication also satisfies the society’s basic needs for rapid and convenient mobile communication. However, due to constraints such as models and manufacturers, mobile terminals are bound to have limited processing power, memory capacity, network connection, and battery capacity, etc. Despite the greatly increased processing speed, storage capacity, and battery capacity of mobile terminals in recent years, network resource allocation of the same generation has inconsistent resource allocation and scheduling due to different mobile phone models, network speeds, paging strategies, etc., leading to information delays under the same network in the same region. At present, similar resource scheduling configurations pay more attention to energy-conscious resource allocation and uncertain energy resource allocation, but fail to comprehensively consider the dynamics of mobile terminals and the difference in user requested resources. Thus, considering uncertainty in paging scheduling under mobile cloud resource environment from the perspective of consumers has become the key to solving the problem of resource allocation in the mobile cloud computing environment.

On the other hand, mobile cloud computing involves the definition of QoS attributes and the impact on mobile terminal performance. In terms of the definition of QoS attributes, literature [5] classifies QoS attributes into five aspects: service response time, cost, reliability, availability and reputation. Literature [6] classifies QoS attributes into five aspects: performance, robustness, security, reputation and others. In the research on resource allocation algorithms, literature [7] proposes genetic algorithms in a virtualized environment, but genetic algorithms usually require a long execution time; literature [8] focuses on negotiation to meet customer SLA requirements; literature [9] proposes virtual machine-based service priority resource allocation scheme. The above research schemes have the following deficiencies: 1) The impact of QoS attributes on mobile terminal performance is not considered; 2) The relationship between weights and node attributes is not considered during matching; 3) In resource allocation, the number of virtual machines used and the use of minimum free space of virtual machines are not considered.

In view of the above deficiencies, our research mainly considers uncertainty of paging scheduling in mobile cloud resource environment under the same mobile network (such as 4 G communication network), but different dynamics of mobile terminals and user requested resources from the consumer’s perspective, so that its delay is minimized and the paging strategy is optimized. In this way, users in the same area can access mobile network communications at the same speed under different terminal devices. This paper proposes an adaptive matching resource allocation algorithm based on uncertain factors under mobile cloud computing environment. The uncertain factors of the mobile terminal are gathered via QoS attribute, and then load characteristics of the user requested resources are analyzed through CLIQUE similarity matching. The mapping between similarity and resources is based on adaptive matching resource allocation algorithm for resource paging allocation.

Our main contributions are manifested as follows:

(1) The uncertain factors that affect the mobile network communication speed are determined through QoS, and user information is collected in real time to solve the inconsistency of user access data in the same area due to the data delay, so that the delay is minimized;

(2) The network resource loading is optimized based on CLIQUE. The analysis and research on the characteristics of user network resources is mainly to analyze load characteristics of real network traffic so that its terminal load resources are maximized;

(3) Mobile communication network paging strategy is proposed based on adaptive matching resource allocation algorithm to ensure that users are not restricted by mobile terminal models, paging strategy is optimized to ensure optimal network speed.

2 Solution

In some applications of mobile cloud computing, there is quite frequent data interaction, and some service transmission involves big data volume in examples of 3D videos, 3D games, website picture groups, etc. Therefore, CLIQUE algorithm is proposed to determine the uncertain factors that affect mobile network communication speed via QoS, acquire user information in real time, cache some or all of the specific cloud service. In this way, users can directly access the needed information, or make some personalized service settings to reduce or even eliminate delays. The network resource loading is optimized based on CLIQUE. The analysis and research on the characteristics of user network resources is mainly to analyze the load characteristics of real network traffic, so that the terminal load resources are maximized. A mobile communication network paging strategy is proposed based on adaptive matching resource allocation algorithm to ensure that users are not restricted by mobile terminal models, paging strategy is optimized to ensure optimal network speed.

2.1 Acquisition of user information in real time based on CLIQUE algorithm

In mobile cloud computing, due to the terminal’s own characteristics (diversity, mobility, robustness, etc.), the network access is uneven. Plus different conditions of the mobile terminal, Q_oS attributes are very closely related to the mobile terminal. Literature [10] believes that the definition of Q_oS attributes includes CPU usage (C_cpu), which refers to the machine’s runnable program conditions at a certain time point; battery availability (C_bat), the available power of the mobile terminal battery; signal strength (C_loss), which refers to the ratio of the number of lost data packets to the transmitted data packets; packet loss rate (C_sign), which is used to judge the communication quality; round trip delay (C_rtt), which indicates the delay experienced since the sender requests resources, etc. Assume that Q_u represents the actual CPU usage rate of the user, Q_t represents the user’s actual battery availability, Q_s represents the user’s actual packet loss rate, Q_l represents the user’s actual signal strength, Q_i represents the round-trip delay for the user to request resources, and Q_r represents performance of the user requested resource, Q_m represents the terminal performance, Q_t represents the requested resource size, Q_c represents computing performance, and Q_n represents network performance. According to the above description, when seeking resource scheduling based on CLIQUE, the user information matching p of the Q_oS attribute tree is acquired by formula: $\begin{matrix} p = 1 - {[(\frac{C_{cpu} - C_{{cpu}^{'}}}{C_{cpu}} Q_{u} + \frac{C_{bat} - C_{{bat}^{'}}}{C_{bat}} Q_{b}) Q_{c} + \\ (\frac{C_{sign} - C_{{sign}^{'}}}{C_{sign}} Q_{s} + \frac{C_{loss} - C_{{loss}^{'}}}{C_{loss}} Q_{l} + \frac{C_{rtt} - C_{{rtt}^{'}}}{C_{rtt}} Q_{t})] Q_{c} \\ + \frac{C_{tasks} - C_{{tasks}^{'}}}{C_{tasks}} Q_{n}} \end{matrix}$ (1)

According to the above description, after knowing that user information is matched, characteristics space in the l-dimensional space with user area side length of §can be determined based on CLIQUE (Clustering In QUEst). Each unit u is written as u_t1×u_t2×...u_tk (t₁ _< t₂...<t_k, k≦l), and there are three main stages. The first stage is to define the subspace where the user’s mobile network occurs. For the specific implementation: first identify all k-dimensional density units (1≦k≦l), and then select the subspaces containing user density units, identify all one-dimensional density units along the user feature space. In each step, determine the k-dimensional density unit set Dk based on the (k-1)-dimensional density unit set Dk-1. After the process is completed, we consider all the subspaces containing density units, and classify the subspaces according to their coverage, namely the point fragments of the original data set in the user’s area. The second stage is CLIQUE-based cluster identification. The user subspace with high coverage determined in the previous step is used as the input of this stage. Each subspace forms a cluster. Arbitrarily select one such unit, and identify all the units connected to it, so that they are assigned to a new cluster C and a second cluster is generated in the same process, and so on. The purpose of the third stage is to seek a minimum cluster description for each cluster, which is divided into two steps. The first step is to randomly select a cluster C density unit formed in the previous stage, extend it in two directions in one dimension to cover as many units in C as possible. Then, the unit is extended in the second direction (as in the previous case) until all dimensions are considered. The second step is to consider all the covered areas to generate each cluster, and delete those units covered by another area. Therefore, assume that the user uses a k-dimensional network, then the user area coverage value is: $H (X^{k}) = - \sum_{i = 1}^{n} c p_{i} l Q_{t} \int_{l}^{1} (Q_{c} - Q_{n})$ (2)

Based on (1) and (2), the real-time acquisition information of the user is measured: $X^{k} = \sum_{j = 1}^{k} H (x_{t j}^{k}) - H (X^{k})$ (3)

2.2 CLIQUE-based network resource loading optimization

Due to the limited mobile terminal resources, the most effective way to improve data processing capabilities is to select remote cloud service access. Due to the constraints of mobile terminal energy consumption and business delays, mobile terminal users face defects such as calculation delay and information cloud residence time when receiving information. Based on the above-mentioned user real-time information acquisition parameters, the relative load value L[i] is calculated. L[i] must be able to reflect the performance difference between resources of different models and the potential load intensity, thus providing the best decision-making basis to achieve network load balancing of different models. The formula of L[i] is defined as follows: $L [i] = {\begin{matrix} 1, i f (U r_{i} = U R_{i} o r U h_{i} \geq U H_{i} o r U q_{i} \geq U Q_{i}) \\ w_{1} \frac{U r_{1}}{U R_{1}} + w_{2} \frac{U r_{2}}{U R_{2}} + w_{3} \frac{U r_{3}}{U R_{3}}, O t h e r s \end{matrix}$ (4)

Where, Ur_i and UR_i respectively represent the current resource request amount and the maximum request amount; Uh_i, UH_i represent the current service capacity and the maximum service capacity; Uq_i and UQ_i represent the current service intensity and the maximum service intensity.

Then, according to the value of (4) and formulas (1) and (2), the maximum network resource load (5) can be obtained as: $Max = \sum_{i}^{1} Q_{n}^{a} - (Q_{n}^{d} + Q_{n}^{c})$ (5)

Where: $\begin{matrix} Q_{n}^{d} = δ_{n}^{d} • c_{n}^{net} \end{matrix}$ (6) $Q_{n}^{c} = U_{n} • δ_{n}^{s} • C_{n}^{c}$ (7)

2.3 Mobile communication network paging strategy based on adaptive matching resource allocation algorithm

Under the reliable location observation information of the host, as for high probability location distribution of the host, the optimal paging area size of the host at a certain moment can be known based on time interval-based one-step paging strategy to allow minimum paging cost. However, this strategy is greatly affected by various host parameters (such as speed, time, number of sampling points, and confidence parameters, etc.). Parameter changes will greatly affect the prediction results, which makes it difficult to guarantee the corresponding prediction results. How to choose the corresponding mobile model to reduce parameter effects and guarantee the prediction effect should be an important aspect concerned by researchers. Therefore, this paper describes the user information and network traffic volume maximization, and proposes adaptive paging strategy based on delay restriction paging, with the basic idea as follows: Suppose there are N mobile devices in a certain area, which are represented by U₁,U₂...U_n and recorded as Ω = {U₁,U₂...U_n}. Use Pt(Ci) to indicate the probability of paging in the area Ci at t-th incoming call. When the t-th incoming call reaches the required area, according to formula (1) and (5), group all mobile equipment selection information resource p in Ω in descending order, and continuously update according to the resource allocation size: $p_{t} (c_{i}) = p_{t - 1} (Max) + p (1 - Max), (c \in Ω, c \neq c_{i})$ (8)

For example, when the t-th incoming call arrives in the mobile area, the following operations will be performed:

(1) t = t+1,s = 0

(2) Arrange all mobile devices in the set Ω according to the resource request information p in descending order, and divide them into n groups;

(3) s = s+1, try to find all paging information on the mobile terminal, if so, go to (4), otherwise go to (3);

(4) Update the found call request to guarantee synchronization of all mobile terminals;

(5) Update the resource allocation information probability of the t-th call in this area;

(6) End.

3 Simulation experiment and result analysis

To verify the CLIQUE-based resource allocation algorithm in a mobile environment, this paper uses CloudSim simulation platform to compare experiments, and experimentally verifies the proposed adaptive algorithm strategy at the same time.

3.1 Test bench and research methods

Based on the CloudSim simulation platform, the virtual machine method adopted by the original service provider is compared with this algorithm, and the experimental results are analyzed.

The data used herein is shown in Table 1, and the number of mobile terminals is m = 10. To simulate the reality as much as possible without loss of generality, the QoS attribute values in the mobile terminals are randomly generated, and parameters like CPU, memory, network, battery capacity, throughput, voltage, load completion time are tested as sample test data. The experimental results are shown in Table 1.

Table 1
Heterogeneity of Computing Devices in the Testbed

Samsung GALAXY Note 4 HUAWE honor 6 Plus iPhone 6 A1586 MI 5 OPPO A53 ZTE Lenovo

CPU Qualcomm Quad core 2.7GHz Hisilicon Kirin 925 Eight core 1.8GHz Dual-core A8 1.4GHz Qualcomm 820 1.8GHz Qualcomm MSM8939 1.5GHz MT6735 Quad core 1.3GHz Qualcomm Snapdragon808 1.8GHz

Memory(RAM) 3GB ARM 3GB RAM 1GB RAM 3GB RAM 2GB RAM 2GB RAM 3GB RAM

Network 4G,3G,2G, 802.11b/g/n 4G,3G,2G, 802.11b/g/n 4G,3G,2G, 802.11b/g/n 4G,3G,2G, 802.11b/g/n 4G,3G,2G, 802.11b/g/n 4G,3G,2G, 802.11b/g/n 4G,3G,2G, 802.11b/g/n

Battery capacity 3000mAh 3600mAh 1810mAh 3000mAh 3075mAh 2800mAh 3600mAh

Battery voltage 5V 5V 3.85V 5V 5V 5V 5V

Workload completion time 38s 38s 37s 38s 38s 38s 38s

Samsung GALAXY Note 4	HUAWE honor 6 Plus	iPhone 6 A1586	MI 5	OPPO A53	ZTE	Lenovo
CPU	Qualcomm Quad core 2.7GHz	Hisilicon Kirin 925 Eight core 1.8GHz	Dual-core A8 1.4GHz	Qualcomm 820 1.8GHz	Qualcomm MSM8939 1.5GHz	MT6735 Quad core 1.3GHz	Qualcomm Snapdragon808 1.8GHz
Memory(RAM)	3GB ARM	3GB RAM	1GB RAM	3GB RAM	2GB RAM	2GB RAM	3GB RAM
Network	4G,3G,2G, 802.11b/g/n	4G,3G,2G, 802.11b/g/n	4G,3G,2G, 802.11b/g/n	4G,3G,2G, 802.11b/g/n	4G,3G,2G, 802.11b/g/n	4G,3G,2G, 802.11b/g/n	4G,3G,2G, 802.11b/g/n
Battery capacity	3000mAh	3600mAh	1810mAh	3000mAh	3075mAh	2800mAh	3600mAh
Battery voltage	5V	5V	3.85V	5V	5V	5V	5V
Workload completion time	38s	38s	37s	38s	38s	38s	38s

3.2 Adaptive test

Time delay and cost test are carried out using adaptive paging and basic paging strategies through five devices. In the simulation experiment, assume that the incoming call arrival rate of the mobile terminal is β, and the stay time of the mobile terminal in each area obeys the general distribution, with an average value of 1/α. Then, the incoming mobile ratio CMR of each terminal is β/α. Take β= 0.02, α= 0.9, n = 4, r = 10, and take CMR = 0. 01,0.1,1,10, calculate the average of 200 runs. Figures 1 –4 can thus be obtained. The analysis result shows that Figs. 1 and 2 respectively reflect the cost to delay ratio of “adaptive paging” and “basic paging strategy”.

Fig. 1

Cost ratio between “adaptive paging” and “basic paging strategy”.

From Figs. 1 and 2, it can be seen that “adaptive paging” has smaller cost and delay than “basic paging strategy”, which changes with the change of CMR. Figures 3 and 4 respectively reflect the cost to delay ratio of “basic paging” (when paging, regardless of the history record of the area visited by the mobile station, each time it is grouped according to fixed mobile settings, and then paged by group) and “basic paging strategy”. From Figs. 3 and 4, it can be seen that the cost and delay of “basic paging” seems chaotic.

Fig. 2

Delay ratio of adaptive paging.

Fig. 3

The cost ratio of “non-adaptive paging” and “basic paging strategy”.

Fig. 4

The cost ratio of “non-adaptive paging” and “basic paging strategy”.

Through a variety of experiments, it is found from different task modules and different service providers that the method proposed in the literature has minimum delay within the user deadline and optimal load resource ratio, so it is possible that mobile terminals of different models have consistent paging ability in the same mobile network environment.

4 Conclusion

This paper first grasps the uncertain factors of the mobile terminal through the QoS attribute, matches the real-time user information based on the CLIQUE algorithm, and on this basis, realizes optimization of network resource loading. Then, based on the mapping between similarity and resources, resource paging allocation is performed based on self-adaptive matching resource allocation algorithm. The simulation experiment analysis shows that this algorithm can achieve minimum delay and optimal load resource ratio, making it possible to let consumers have the same paging capacity under the same network when mobile terminals are different.

Footnotes

Acknowledgments

Master Program Training Team Building Plan of Jiyang College of Zhejiang A & F University (Jiyang College [2018] No. 112).

References

, Liang

S.H.

, Marchionini

, Mobile Search Behaviors An In-depth Analysis Based on Contexts, APPs, and Devices, Morgan and Claypool eBooks, 2018.

Ali

H.G.E.D.H

, Saroit

I.A.

and Kotb

A.M.

, Grouped tasks scheduling algorithm based on QoS in cloud computing network, Egyptian Informatics Journal18(1) (2017), 11–19.

Zhou

, Li

, Ge

, Li

, Zhou

and Luo

, Multi-Objective Workflow Scheduling Based on Delay Transmission in Mobile Cloud Computing, Journal of Software29(11) (2018), 3306–3325.

Zhang

and Wen

, Energy-Efficient Task Execution for Application as a General Topology in Mobile Cloud Computing, IEEE Trans On Cloud Computing6(3) (2018), 708–719.

Yong

, Chun-Ming

and Zong-Xia

, QoS-Awared Grid Workflow Schedule, Journal of Software17(11) (2006), 2341–2351.

Guangwen

, Design and implementation of UDDI extended model with QoS support and dynamic lease mechanism, Qingdao: Ocean University of China, 2008.

Xia-yu

, Jun

and Wen-xin

, Ant colony optimization algorithm for computing resource allocation based on cloud computing environment, Journal of East China Normal University(Natural Science)1 (2010), 127–134.

Son

, Cluster negotiation-based flexible SLA establishment with SLA-driven resource allocation in cloud computing, IEEE/ACM International Symposium on13 (2013), 168–171.

Cuervo

, Balasubramanian

, Cho

D.-K.

, Wolman

, Saroiu

, Chandra

and Bahl

, MAUI: Making smart-phones last longer with code offload, in Proc. 8th Int. Conf. Mobile Syst., Appl., Serv., San Francisco, CA, USA, 2010, 49–62.

10.

Haiyang,

Liu Runhua

and Hua

, Multi-objective optimization for task scheduling in mobile cloud computing, Journal of Computer Research and Development54(9) (2017), 1909–1919.

11.

M.-R.

, Sheth

, Mummert

, Pillai

, Wetherall

and Govindan

, Odessa: Enabling interactive perception applications on mobile devices, in Proc. 9th Int. Conf. Mobile Syst., Appl., Serv., Bethesda, MD, USA, (2011), 43–56.

12.

Viswanathan

, Lee

E.K.

and Pompili

, An autonomic resource provisioning framework for mobile computing grids, in Proc. 9th Int. Conf. Auton. Comput., San Jose, CA, USA, Sep. (2012), 79–84

13.

Zuyao

and Yueming

, Joint optimization method for multi-workflow task scheduling in mobile cloud computing environment, Application Research of Computers37(8) (2020), 2261–2265.

14.

Song

, Li

, Wang

, et al., A service-oriented priority-based resource scheduling scheme for virtualized utility computing, High Performance Computing-HiPC 2008, 23–26.

15.

Chun

B.-G.

, Ihm

, Maniatis

, Naik

and Patti

, CloneCloud: Elastic execution between mobile device and cloud, in Proc. 6th Conf. Comput. Syst., Salzburg, Austria, (2011), 301–314.

16.

Zhu

Y.H.

and Leung

C.M.

, Optimization of sequential paging in movement-based location management based on movement statistics, IEEE Transaction on Vehicular Technology56(2) (2007), 955–964.

17.

Han

, Harrold

, Armour

, et al., Green radio: radio techniques to enable energy efficient wireless networks, IEEE Communications Magazine49(6) (2011), 46–52.

18.

Viswanathan

, Lee

E.K.

, Rodero

and Pompili

, Uncertainty-Aware Autonomic Resource Provisioning for Mobile Cloud Computing, EEE Transactions on Parallel and Distributed Systems26(8) (2015), 2363–2372.