Mobile-cloud collaborative Convolutional Neural Network (CNN) inference enables the efficient execution of CNN models by offloading partial inference workloads from mobile devices to the cloud. Although model partitioning for collaborative inference has been extensively studied, most existing approaches assume reliable mobile-cloud transmission, which often breaks down in real-world wireless environments with packet loss. In such scenarios, incomplete feature transmission can result in a significant drop in inference accuracy. In this paper, a joint scheduling approach is proposed to address this challenge, leveraging a search-based algorithm to determine both the model partition layer and redundancy level, to balance inference accuracy and latency under packet loss conditions. The proposed method is evaluated in a real-world mobile-cloud environment. Results show that it reduces the inference latency by up to 30.3% compared to the non-redundant baseline with the same accuracy threshold.

Support Vector Machine (SVM) is a machine learning with excellent classification performance, which has been widely used in various fields such as data mining, text classification, face recognition and etc. However, when data volume scales to a certain level, the computational time becomes too long and the efficiency becomes low. To address this issue, we propose a parallel balanced SVM algorithm based on Spark, named PB-SVM, which is optimized on the basis of the traditional Cascade SVM algorithm. PB-SVM contains three parts, i.e., Clustering Equal Division, Balancing Shuffle and Iteration Termination, which solves the problems of data skew of Cascade SVM and the large difference between local support vector and global support vector. We implement PB-SVM in AliCloud Spark distributed cluster with five kinds of public datasets. Our experimental results show that in the two-classification test on the dataset covtype, compared with MLlib-SVM and Cascade SVM on Spark, PB-SVM improves efficiency by 38.9% and 75.4%, and the accuracy is improved by 7.16% and 8.38%. Moreover, in the multi-classification test, compared with Cascade SVM on Spark on the dataset covtype, PB-SVM improves efficiency and accuracy by 94.8% and 18.26% respectively.

The limitation of simulation scales leads to a gap between simulation results and physical phenomena. This paper reports our efforts on increasing the scalability of metal material microscopic damage simulation on the Sunway TaihuLight supercomputer. We use a multiscale modeling approach that couples Molecular Dynamics (MD) with Kinetic Monte Carlo (KMC). According to the characteristics of metal materials, we design a dedicated data structure to record the neighbor atoms for MD, which significantly reduces the memory consumption. Data compaction and double buffer are used to reduce the data transfer overhead between the main memory and the local store. We propose an on-demand communication strategy for KMC to remarkably reduce the communication overhead. We simulate 4 * 1012 atoms on 6,656,000 master+slave cores using MD with 85% parallel efficiency. Using the coupled MD-KMC approach, we simulate 3.2 * 1010 atoms in 19.2 days temporal scale on 6,240,000 master+slave cores with runtime of 8.6 hours.

At present, MapReduce is the most popular programming model for Big Data processing. As a typical open source implementation of MapReduce, Hadoop is divided into map, shuffle, and reduce. In the mapping phase, according to the principle moving computation towards data, the load is basically balanced and network traffic is relatively small. However, shuffle is likely to result in the outburst of network communication. At the same time, reduce without considering data skew will lead to an imbalanced load, and then performance degradation. This paper proposes a Locality-Enhanced Load Balance (LELB) algorithm, and then extends the execution flow of MapReduce to Map, Local reduce, Shuffle and final Reduce (MLSR), and proposes a corresponding MLSR algorithm. Use of the novel algorithms can share the computation of reduce and overlap with shuffle in order to take full advantage of CPU and I/O resources. The actual test results demonstrate that the execution performance using the LELB algorithm and the MLSR algorithm outperforms the execution performance using hadoop by up to 9.2% (for Merge Sort) and 14.4% (for Word Count).

Different from the traditional topic detection and tracking (TDT) technologies primarily focus on detecting topic in local network domain, this paper proposes a cooperative alert topic detection model in distributed environment (named CATDM). The model abstracts the alert topic and represents it as the local alert case by analyzing the alert of campus network culture in depth. The model not only discovers new alert topic of local network domain, but also cooperatively schedules the information of alert case knowledge base (ACKB) between different alert monitor nodes. CATDM discovers new alert topic of each monitor node and optimizes the local ACKB periodically. Through cooperatively scheduling the information of ACKB between different alert monitor nodes, the model enables some alert monitor nodes to obtain the ability of detecting new alert topic and strengthens the detection ability of burst alert topic. To validate the performance of CATDM, we present two comparison experiments on the data corpus about ¿campus network culture¿. The experiment results validate the feasibility and practicality of CATDM, and demonstrate that CATDM can effectively improve the detecting ability of burst alert topics of local network domain in distributed environment.

Aiming at the deficiencies of traditional blog hotness evaluation methods, the paper presents a blog hotness evaluation model based on text opinion analysis (named BHEM-TOA). The model not only considers the number of reviews, comments and publication time of the blog topic, but also focuses on the comment opinion. BHEM-TOA emphasizes subjective opinions of reviewers about the blog topic. It utilizes the text opinion analysis method based on Chinese characters to extract opinioned comments, gets supportive and oppositive circumstances about the blog topic, then combines with the number of reviews, comments and publication time to realize blog hotness evaluation. To validate the performance of BHEM-TOA, the experiment constructs two data corpuses called TOAC and BHEC, and the experimental results demonstrate that BHEM-TOA could more precisely and comprehensively evaluate the hotness of the blog than traditional methods.

In regards to applications like 3D seismic migration, it is quite important to improve the I/O performance within an cluster computing system. Such seismic data processing applications are the I/O intensive applications. For example, large 3D data volume cannot be hold totally in computer memories. Therefore the input data files have to be divided into many fine-grained chunks. Intermediate results are written out at various stages during the execution, and final results are written out by the master process. This paper describes a novel manner for optimizing the parallel I/O data access strategy and load balancing for the above-mentioned particular program model. The optimization, based on the application defined API, reduces the number of I/O operations and communication (as compared to the original model). This is done by forming groups of threads with "group roots", so to speak, that read input data (determined by an index retrieved from the master process) and then send it to their group members. In the original model, each process/thread reads the whole input data and outputs its own results. Moreover the loads are balanced, for the on-line dynamic scheduling of access request to process the migration data. Finally, in the actual performance test, the improvement of performance is often more than 60% by comparison with the original model.

Many important scientific kernels compute solutions using finite difference techniques, and the most time consuming part of them is the iterative method, such as Gauss-Seidel or SOR. To improve performance, iterative method can exploit parallelism, intra-iteration data reuse, and inter-iteration data reuse. This paper describes a new parallel Gauss-Seidel method using iteration space alternate tiling technique, which is developed not only to improve parallelism, intra-iteration, and inter-iteration data locality, but also to decrease communication and synchronization cost in iterative method. Time-skewing can increase cache locality by exploiting locality in the time direction as well as spatial locality. The degree of parallelism can be improved by reordering the execution of cache blocks. Finally numerical results are presented which confirm the effectiveness of Gauss-Seidel parallelized with iteration space alternate tiling technique, specifically compared with owner-computes and red-black coloring based Gauss-Seidel methods, and show that the new method has a good parallel performance on distributed memory machines, as well as scalability.