ABSTRACT. Collecting the trajectories occurring in the city and mining the patterns implied in the trajectories can support the ITS (Intelligent Transportation System) applications and foster the development of smart cities. For improving the operations of inter-zonal buses in the cities, we define a new trip pattern, i.e., frequent bus passenger trip patterns for bus lines (FBPT4BL patterns in short). We utilize the passenger trajectories from bus smart cards and propose a two-phase approach to mine FBPT4BL patterns and recommend inter-zonal bus lines. We conduct extensive experiments on the real data from the Beijing Public Transport Group. By comparing the experimental results with the actual operation of inter-zonal buses at the Beijing Public Transport Group, we verify the validity of our proposed method.

ABSTRACT. With the rapid development of information technology, recommender systems have become critical components to solve information overload. As an important branch, weighted hybrid recommender systems are widely used in electronic commerce sites, social networks and video websites such as Amazon, Facebook and Netflix. In practice, developers typically set a weight for each recommendation algorithm by repeating experiments until obtaining better accuracy. Despite the method could improve accuracy, it overly depends on experience of developers and the improvements are poor. What worse, workload will be heavy if the number of algorithms rises. To further improve performance of recommender systems, we design an optimal hybrid recommender system on Spark. Experimental results show that the system can improve accuracy, reduce execution time and is able to handle large-scale datasets. Accordingly, the hybrid recommender system balances accuracy and execution time.

ABSTRACT. Using multiple streams can improve the overall system performance by mitigating the data transfer overhead on heterogeneous systems. Currently, very few cases have been streamed to demonstrate the streaming performance impact and a systematic investigation of streaming necessity and how-to over a large number of test cases leaves a gap. In this paper, we use a total of 56 benchmarks to build a statistical view of the data transfer overhead, and give an in-depth analysis of the impacting factors. Among the heterogeneous codes, we identify two types of non-streamable codes and three types of streamable codes, for which a steaming approach has been proposed. Our experimental results on the CPU-MIC platform show that, with multiple streams, we can improve the performance by up 90%. Our work can serve as a generic flow of using multiple streams on heterogeneous platforms.

ABSTRACT. GPU-accelerated computing gains rapidly-growing popularity in many areas such as scientific computing, database systems, and cloud environments. However, there are less investigations on the security implications of concurrently running GPU applications. In this paper, we explore security vulnerabilities of CUDA from multiple dimensions. In particular, we first present a study on GPU stack, and reveal that stack overflow of CUDA can affect the execution of other threads by manipulating different memory spaces. Then, we show that the heap of CUDA is organized in a way that allows threads from the same warp or different blocks or even kernels to overwrite each other’s content, which indicates a high risk of corrupting data or steering the execution flow by overwriting function pointers. Furthermore, we verify that integer overflow and function pointer overflow in struct also can be exploited on GPUs. But other attacks against format string and exception handler seems not feasible due to the design choices of CUDA runtime and programming language features. Finally, we propose potential solutions of preventing the presented vulnerabilities for CUDA.

ABSTRACT. As an interesting commercial application, rendering plays an important role in the field of animation and movie production. Generally, render farm is used to rendering mass images concurrently according to the independence among frames. How to scheduling and manage various rendering jobs efficiently is a significant issue for render farm. Therefore, the prediction of rendering time for frames is relevant for scheduling, which offers the reference and basis for scheduling method. In this paper a statistics based prediction method is addressed. Initially, appropriate parameters which affect the rendering time are extracted and analyzed according to parsing blend formatted files which offers a general description for synthetic scene. Then, the sample data are gathered by open source software Blender and J48 classification algorithm is used for predicting rendering time. The experimental results show that the proposed method improve the prediction accuracy about 60% and 75.74% for training set and test set, which provides reasonable basis for scheduling jobs efficiently and saving rendering cost.

ABSTRACT. Software defined network (SDN) is an emerging network architecture that has drawn the attention of academics and industry in recent years. The key idea of SDN is separating control plane and forwarding plane to simplify the management of network, the deployment of application and so on. Affected by investment protection, risk control and other factors, the full deployment of SDN will not be finished in the short term, thus it results into a coexistence state of traditional IP network and SDN which is named hybrid SDN. The exiting hybrid SDN scenario calculates and only uses a single shortest path to optimize the traffic passing through SDN nodes. In this paper, we formulate the SDN controller’s optimization problem for load balancing as a mathematical model. Then we propose a routing algorithm Dijkstra-Repeat in SDN nodes which can offer disjoint multipath routing. To make it computationally feasible for large scale networks, we develop a new Fast Fully Polynomial Time Approximation Schemes (FPTAS) based Lazy Routing Update (LRU). Theoretical analysis and simulations validate the efficacy of our algorithms.

ABSTRACT. Abstract. Recently, Deep Learning (DL) has become super hot because it achieves breakthroughs in many areas such as image processing and face identication when there are enough data. The performance of DL models critically depend on hyperparameter settings. However, it is extremely dicult to manually tune these parameter settings without indepth knowledge about the DL algorithm and its input data. Although the recently proposed Bayesian optimization methods have achieved considerable successes in optimizing these hyperparameter settings and even surpass the performance of human experts, blind reliance on such approaches makes how these hyperparameters take eects unknown to end users. To address this issue, a random-forest based approach is proposed to quantify the importance of the hyperparameters of DL models in terms of performance. Nevertheless, this approach needs a long time to collect a large amount of data for training the random forest models, which hampers users to use it in practice.

ABSTRACT. Crowd Sensing between users with smart mobile devices is a new trend of development in Internet. In order to recommend the suitable service providers for crowd sensing requests, this paper presents a Fine-grained Crowdsourcing Model (FCM) based on Ontology theory that helps users to select appropriate service providers. First, the characteristic properties which extracted from the service request will be compared with the service provider based on ontology triple. Second, recommendation index of each service provider is calculated through similarity analysis and cluster analysis. Finally, the service decision tree is proposed to predict and recommend appropriate candidate users to participate in crowd sensing service. Experimental results show that this method provides more accurate recommendation than present recommendation systems and consumes less time to find the service provider through clustering algorithm.

ABSTRACT. In this paper, we studies the problem to find the maximal number of red nodes of a kind of balanced binary search tree. We have presented a dynamic programming formula for computing $r(n)$, the maximal number of red nodes of a red-black tree with $n$ keys. The first dynamic programming algorithm uses $O(n^2\log n)$ time and uses $O(n\log n)$ space. The basic algorithm is then improved to a more efficient $O(n)$ time algorithm. The time complexity of the new algorithm is finally reduced to $O(n)$ and the space is reduced to only $O(\log n)$.

ABSTRACT. HTCondor, a batch system characterized by its matchmaking mechanism, schedules job in FCFS way, so its performance is not ideal as expected. Backfilling is a technique to address the above problem which schedules low-priority small jobs before currently blocked jobs to fill the resource gap. Most backfilling algorithms are based on CPU information and have large room for improvements with considering other resource information. The K-resource aware scheduling algorithm Backfill Balanced (BB) selects backfill job which can best balance the usage of all resources and achieve better performance compared with the classical backfilling algorithm. However, BB does not realize that small jobs’ impacts on resource utilization are negligible and they mainly contribute to reduce the average response time. Here we propose the IBB algorithm, which utilizes the characteristics of small jobs to guide a better job selection. We implemented IBB on HTCondor to improve its performance. Experiments results show that IBB can provide up to 60% performance gains in most performance metrics compared with BB.