Research On Storage Infrastructure Of Ad-Hoc Mobile Cloud

With the development of mobile network and Internet of Things(IoT),we are moving gradually from big data era to mobile big data era.In statistic,the total number of mobile devices will increase to 2.8 billion by 2020,and corresponding mobile data scale are increasing with 53%per year.Meanwhile,the proportion of mobile data in global are increasing too,and it will become the biggest part of big data.To alleviate the contradiction between high data scale increasing speed and limited hardware resource of mobile devices,many researchers from both academic and industry domain are studying to efficiently organize mobile cloud any time and any where,so that the large scale of mobile data can be dealt with in mobile side.This logic model is called ad-hoc mobile cloud.Efficiently organizing and managing mobile data is important,because it is the fundamental to guarantee the QoS of ad-hoc mobile cloud.Unfortunately,the study on storage infrastructure of ad-hoc mobile cloud is just beginning,many related works just port techniques from traditional cloud storage and cannot make full use of the compli-cate hardware as well as software environment characteristics.The characteristics in-cluding:(1)The hardware environment between mobile devices and traditional server is totally different,for example,the Flash medium is more good at random writing and reading,while the bandwidth of wireless communication is much lower than wire net-work;(2)Most of the mobile data is structured or semi-structured,and it is scattering,scale-efficient and requires high demand on security;(3)Mobile users present higher demand on real time processing and data security.In this paper,we analyze the characteristic of ad-hoc mobile cloud carefully,and further propose a novel storage infrastructure with higher "efficiency" and "security''.The main work and contributions are as follows:1.Optimization of Hadoop on local mobile devicesAs the first step of our work,we port Hadoop to mobile side successfully.Dur-ing the process,we find that the performance of' traditional cloud frameworks performs not very good.To address the issue,we propose a novel optimization scheme towards Hadoop porting.With the help of neutral network,the internal relationship between configuration parameters and hardware/software environment in mobile side is dug out.The sense of this work is:first of all,we found a way to port Hadoop efficiently and achieved it,which is the fundamental of our following researches;Second,we learn more knowledge about the performance and bottleneck of data stored in mobile side.Lastly,experimental results show that porting Hadoop on mobile devices straightfor-ward is not a good way,the purpose on "efficiency" and "security" cannot be achieved in this way.2.Light weight key-value store on local distributed mobile devicesMobile data are always structured or semi-structured,which cannot be organized very well by distributed file systems.Based on earlier works,we propose a light weight distributed key-value storage system on local mobile devices,called LKSM.The system is indexed by LSM-Tree,and optimized by resilient DRAM slot.With the cooperation of above two components,LKSM addressed the issue of writing sensitive and I/O ampli-fication.Meanwhile,the storage performance is significantly improved.Experimental results show that LKSM outperforms LevelDB with 30.12%and 36.16%of read/write overhead(data size = 512MB).3.Attribute-namespace based security model for ad-hoc mobile cloudEarlier works are mainly focus on the efficiency of storage infrastructure,however,since mobile data has high relationship with people's privacy,mobile users require high demand on security.In this paper,we analyze the storage model with three layers:user layer,logical layer and physical layer.We find the weakness and propose an attribute namespace based security model based on that.The model is called ANDS.By organiz-ing the attribute set on namespace tree,ANDS achieves the protection of both content data and meta data by access control and data encryption.Furthermore,we propose chunk grained proxy-encryption mechanism,based on this technique,the computation overhead is significantly decreased.4.Self-similarity based load balancing for large scale edge computingIt is very easy to organize a mobile cloud in LAN when the scale of mobile devices is small,however,the amount of mobile devices are much more than PC servers.Thus,the scale of mobile cluster has potential to be very large.With the development of techniques such as SDN and NFV,the number of edge devices is even bigger.As a result,how to organize mobile cluster with large scale is becoming a big issue to address.In this paper,we propose a self-similarity based load balancing for large scale edge computing(SSLB),and further propose two algorithms:Task Distributing and Task Grasping.By analysis and simulation,we proves the efficiency of SSLB successfully.