The default Horizontal Pod Autoscaler (HPA) in Kubernetes is inherently reactive, often taking 15–30 seconds to respond to workload changes on resource-constrained edge devices. In latency-sensitive applications such as real-time crowd analytics, these delays frequently cause thermal throttling and QoS violations during sudden surges in crowd density. To overcome this limitation, we propose Edge-VisForeCast, a proactive autoscaling framework that leverages YOLOv11-detected crowd density as a Granger-causal leading indicator of future resource demand. By integrating these application-level insights with a quantized Informer transformer model enhanced by a Vision-Aware Attention mechanism, the system can forecast power consumption 5–15 seconds ahead. Granger causality tests (p< 0.001) confirm that person count statistically precedes CPU and power saturation by 10–15 seconds. We deployed the proposed system on a 3-node NVIDIA Jetson Orin Nano cluster at the Tunghai University Library, using a real-world dataset of 39.01 million raw frames collected over 14 days. Experimental results show that Edge-VisForeCast achieves 17.8% better forecasting accuracy (MAE) than LSTM baselines, while delivering a 23.7% reduction in inference latency and a 36.3% reduction in energy consumption. Most importantly, the system maintains stable operation within the 15W thermal envelope, effectively avoiding the throttling issues commonly observed with reactive HPA. This work demonstrates the value of bridging application semantics with system orchestration for energy-efficient predictive autoscaling in IoT edge environments.

Identifying pavement damage is a crucial component in road maintenance and infrastructure management. Prompt detection and corrective action of pavement defects can prevent severe deterioration, maintain safety, and prolong the useful life of road infrastructure. Computer vision is widely applied to vehicle applications, such as driver assistance and self-driving, and that is a good platform to equip the pavement damage recognition service. Therefore, this work applies computer vision to develop the pavement damage detection service named high-performance pavement health recognition (HPPH) to detect common pavement damages, including longitudinal cracks, transverse cracks, alligator cracks, and potholes. The proposed HPPH can be deployed on a vehicle for regular road detection. To fit the requirements of vehicle applications, HPPH considers state-of-the-art techniques to optimize the performance of the recognition service on vehicles, e.g., DeepStream is applied to increase the inference performance. At the same time, YOLOv8s (you only look once, YOLO) provides real-time inference. The experimental results reveal that HPPH provides an accuracy of 90.3% and an average precision 0.5 of 74.3%. Moreover, the optimized HPPH provides three times better than the pure YOLOv8s in terms of the inference speed. In summary, the proposed HPPH provides high recognition accuracy with high efficiency and is feasible to be applied to vehicle applications to realize distributed road maintenance.

Location-based Systems (LBS) are popular for delivering customized information. However, some issues, such as place credibility, the efficiency of position calculations, and communication latency, pose challenges for indoor LBS. This work proposes the High-performance Perspective Platform (H3P) to help network managers understand network users’ information. The H3P provides indoor positioning services based on Wi-Fi 6 (IEEE 802.11ax) to stabilize service quality and ensure high computation efficiency for rapid service response. It utilizes Apache Kafka and Apache Zookeeper clusters on Kubernetes to handle large amounts of data. Wi-Fi usage data is transmitted to Kafka’s distributed real-time data streaming to enhance position credibility and the immediacy of position calculations. The data structure is also optimized to improve computation efficiency. Experimental results show that H3P improves data latency by up to 69% and data insertion latency by up to 73%. Additionally, H3P offers more stability and efficiency than Chang et al., 2012 in terms of data transmission, with an improvement of approximately 16.15%. This allows administrators to manage the network with user-friendly interfaces and a smooth user experience.

Wireless network stability is critical for organizations. Most companies rely on a solid Internet connection for at least part of their day-to-day activities. It is essential to show the fast and high capacity of Wi-Fi. Wireless network management also significantly supplies the user with quality service and helps the system administrator manage and track the network infrastructure. The user must be able to stay connected. The Wi-Fi connection engine was applied in this article using the analysis and locations engine (ALE). ALE provides the information, including MAC address, location, floor information, and the building where the device belongs. From this data, we presented the Web visualization of the Wi-Fi network monitoring system on the map using Cesium. Except for the common information about Wi-Fi users, this work presents the vertical position visualization in the buildings to provide more details about user usage. This work analyzes the user experience of using the proposed system, and latency is the appropriate metric to measure the user experience. The experiment results show that the optimized number of cluster partitions is four, and the latency is less than 1.3 s for 4-MB data. Therefore, the system provides a visualization platform with acceptable system response, and Internet managers would consider the proposed system as the management tool.

Intelligent systems influence many aspects of daily life. With the emergence of the Internet of Things (IoT), artificial intelligence (AI), and machine learning (ML), opportunities have been created for smart computing infrastructure. However, problems might arise from the lack of interconnectivity, higher reliability, real-time predictive analytics, and low-latency requirements. Therefore, in this article, we propose the sensors, edge, and cloud (iSEC) framework. The project deploys a smart cloud edge-computing architecture to provide ML and deep learning in the cloud edge environment. Two pilot projects of air quality monitoring system and object detection are demonstrated to evaluate the iSEC framework.

This work's contributions include three innovative concepts, an improved model, two‐stage Lagrange principle, and minimum‐energy scaling optimisation, for quantisation audio watermarking in the wavelet domain. First, discrete wavelet transform (DWT) multi‐coefficients quantisation, composed of arbitrary scaling on the lowest DWT coefficients, and the group‐based signal‐to‐noise ratio (SNR) of these coefficients is connected in a model. Then, the two‐stage Lagrange principle and minimum‐energy approach play two essential roles to obtain the optimal scaling factors. With the proposed scheme, the best fidelity and robustness of embedded audio can be attained and the perceptual evaluation of audio quality (PEAQ) test with an illustration of the relationship between SNR and PEAQ is also performed as well. Simulation results show that each watermarked audio by the proposed method attains a high SNR, good PEAQ, and a low bit error rate (BER). The SNR of most watermarked audios in their method is above 35 or even above 40 and the corresponding subjective difference grade of PEAQ is close to 0. In terms of comparing BER, most of their BER is as low as 2% or less indicating better robustness against many attacks, such as re‐sampling, amplitude scaling, and mp3 compression.

Traffic conditions are among the issues most concerned with the general public, and the freeway is a large-scale Internet-of-Things application. In addition to obtaining real-time road usage information, analysis of local road usage habits is crucial in evaluations of government policy implementation. Using road usage data provided by the electronic toll collection (ETC) system, we investigated the data on road usage history on the freeways. The ELK stack was employed to construct a platform for visualizing real-time road usage information and history in this article; the platform is named the local transportation knowledge (LTK) platform. By analyzing more than 500 million pieces of data, the LTK platform proposed in this article efficiently visualized road usage data and facilitated the acquirement of local road usage knowledge. We verified that residents of other counties and cities commuted to Taichung each day. We also discovered that a considerable number of Taichung city residents were employed in Hsinchu Science Park and commuted between the two places. The LTK platform can present real-time freeway traffic conditions, facilitate in-depth analysis of local road usage data, and provide data to verify the relevant information.

Abstract The influenza problem has always been an important global issue. It not only affects people’s health problems but is also an essential topic of governments and health care facilities. Early prediction and response is the most effective control method for flu epidemics. It can effectively predict the influenza-like illness morbidity, and provide reliable information to the relevant facilities. For social facilities, it is possible to strengthen epidemic prevention and care for highly sick groups. It can also be used as a reminder for the public. This study collects information on the influenza-like illness emergency department visits to the Taiwan Centers for Disease Control, and the PM2.5 open-source data from the Taiwan Environmental Protection Administration's air quality monitoring network. By using deep learning techniques, the relevance of short-term estimates and the outbreak calculation method can be determined. The techniques are published by the WHO to determine whether the influenza-like illness situation is still in a stage of reasonable control. Finally, historical data and future forecasted data are integrated on the web page for visual presentation, to show the actual regional air quality situation and influenza-like illness data and to predict whether there is an outbreak of influenza in the region.

With the flourishing of Internet of Things (IoT) technology, ubiquitous power data can be linked to the Internet and be analyzed for real-time monitoring requirement. Numerous power data would be accumulated to even Terabyte level as the time goes. To approach a real-time power monitoring platform on them, an efficient and novel implementation techniques has been developed and formed to be the kernel material of this thesis. The generic power-data source is provided by the so-called smart meters equipped inside factories located in an enterprise practically. The data collection and storage are handled by the Hadoop subsystem and the data ingestion to Hive data warehouse is conducted by the Spark unit. On the aspect of system verification, under single record query, these software modules: Hive, Spark, and Impala had been tested in terms of query-response efficiency. And for the performance exploration on the statistical query function and data ETL processing. The kernel contributions of this research work can be highlighted by two parts: (1) Multi-layer software modules are adopted to design and implement the real-time power-monitoring platform embedded with some excellent characteristics of high efficiency, high feasibility and low cost. (2) The rudimental experiments are conducted to verify the query-response efficiency, and performance evaluations for the proposed real-time power monitoring platform, which reveals the high feasibility for the target research goals.

In recent years, high availability shared storage will become a popular information technology industry development orientation. Currently, information technology industries emphasize to reduce high risk data requirements and improve read and write performance of data storage. Therefore, the main purpose of this work is to improve read and write performance with the best way on Ceph Storage Cluster. In this system, the data is stored on Hadoop Distributed File System (HDFS), and the data stored in-memory virtual distributed store system that mentioned as Alluxio automatically. Then, the data would be processed through Hadoop Map Reduce method and the output would be inserted into Hadoop Distributed File System and Alluxio environment. The first experiment is to use S3 as application program interface that will connect to RADOS Gateway stored data into Object Storage Daemon (OSD). The second experiment is based on the first out experiment would be through Ceph File System (CephFS) connected to Object Storage Daemon directly. The data is saved in Ceph environment more secure than in Alluxio as in-memory storage system because OSD can be used for data backup based on object storage level. We can use S3 browser (GUI) to maintain data like grant access, maintain folders maintenance, create user accounts, move data location etc. The last one, we used Inkscope monitors all system. If there is any problem, system will give warning or error responds to users automatically.

The present study utilizes VirtualBox virtual environment technology to develop the Personal, small size, Big Data platform that can effectively replicate a VM Hadoop system and provides an environment for developers to easily design and implement Hadoop Map/Reduce programming. This study also performs the benchmark by using the VM Hadoop, small-cluster Hadoop, and NCHC's large-scale Hadoop cluster, Braavos. The benchmark results show that the VM Hadoop is an ideal platform for the Map/Reduce code development and testing purpose, and the Braavos Hadoop cluster is the most appropriate for production runs. Moreover, based on the standard WordCount example, the computing time of Braavos Hadoop cluster is 232 times faster than the small-cluster Hadoop. In addition, an engineering example, the image recognition of flow monitoring, is given to illustrate the way of big image data analytics in the Hadoop system. Finally, the VM Hadoop, in term of a Big Data development platform, is ready for users to download. The first author of this paper would like to give a demonstration for the proposed VM Hadoop system as well as an engineering application.

In order to improve accessibility and efficiency of a cloud system, this work proposed a mechanism to integrate Ceph, HDFS and Swift based on the OpenStack. We first build a heterogeneous storage environment including Ceph, HDFS and Swift based on the open source OpenStack and then measure their performances. To integrate storage services of Ceph, HDFS and Swift, we propose a proportion-based file distribution mechanism. The proportion for file partition is dependent on the remaining storage capacity so that we can distribute those sub files to different storage. This mechanism also enhances the file security. In addition, a high usability user interface is provided so as to make the proposed system more friendly. Experimental results show the efficiency of our system.

This work implement a cloud infrastructure that can monitor the status of OpenStack and monitor the real-time status of virtual machine on OpenStack then achieve to energy saving through live migration. The projects of monitoring include the utilization of CPU, load of memory, and power consumption. These data show in real-time, completely monitor the real-time status of physical machines and virtual machines. It also record the utilization and power consumption of physical machines then show on this cloud infrastructure, to provide experimental evidence for the user as a reference. Base on the power consumption we monitoring, we can automatically allocate virtual machines on every physical machines by live migration, to balance the power consumption of every physical machines. Its not only can avoid idle and waste of resources but also can avoid reducing machine life because of the physical machines always keep in high usage, and achieve to power saving.

With the development of science and technology and lifestyle changes, people's idea focuses on improving the health environment by using Information Technology to provide warning and prediction. To achieve this goal, the proposed Intelligent Indoor Environment Monitoring System in Cloud (iDEMS) combined environmental sensors with ZigBee wireless sensor network technology to store and process environmental data in HBase. The environmental data collected by sensors will be stored and processed cloudy in HBase which support large amounts of data to store in, free to increase storage space easily. It also can compute through Hadoop MapReduce for HBase database to do distributed computing or cloud computing to process environments records.

Summary In the past, the tenacious semiconductor problems of operating temperature and power consumption limited the performance growth for single‐core microprocessors. Microprocessor vendors hence adopt the multicore chip organizations with parallel processing because the new technology promises faster and lower power needed. In a short time, this trend floods first the development of CPU, then also the other peripherals like GPU. Modern GPUs are very efficient in manipulating computer graphics, and their highly parallel structure makes them even more effective than general‐purpose CPUs for a range of graphical complex algorithms. However, technology of multicore processor brought revolution and unavoidable collision to the programming personnel. Multicore processor has high performance; however, parallel processing brings not only the opportunity but also a challenge. The issue of efficiency and the way how programmer or compiler parallelizes the software explicitly are the keys that enhance the performance on multicore chip. In this paper, we propose a parallel programming approach using hybrid CUDA, OpenMP, and MPI programming. There would be two verificational experiments presented in the paper. In the first, we would verify the availability and correctness of the auto‐parallel tools, and discuss the performance issues on CPU, GPU, and embedded system. In the second, we would verify how the hybrid programming could surely improve performance. Copyright © 2016 John Wiley & Sons, Ltd.

In this paper, we use a wildcard mask to implement the load balance method directly on switches or routers and add a user prediction mechanism to dynamically change the range of the wildcard mask, in this way, the load balance mechanism can be applied conforming to real service situations. In our experiment, we tested the accuracies of flow prediction with different prediction algorithms and compared the delay times and balance effects of the proposed method with other load balancers. With the popularity of cloud computing, the demand of cloud infrastructure also increases. As a result, we also applied our load balance mechanism on cloud services and have proven that the proposed method can be easily applied on varieties of service platforms.

The improvement of sanitary conditions and changes of life style making people's health awareness began to rise. The progress of medical and the promotion of medicine and health, the concept of preventive healthcare is widespread. People begin to concern about the quality of the environment where they stayed. The environmental comfort index include temperature, related humidity, illumination, noise and carbon dioxide. Among these, we monitor and record the values of temperature, related humidity and carbon dioxide. In this paper, we built an environment quality monitoring system, which can adjust the air quality in door and monitor the concentration of harmful gases like formaldehyde, volatile organic compounds and carbon monoxide. If the environment comfort value is not up to standard, the related equipment will be turned on. The system will alert if the content of harmful gases exceed the standard. We hope that based on these real-time data, the proposed system can help people make right and timely decisions, and act on time to maintain a beneficial environment in the monitored area.

Currently, most health care systems used among divisions in medical centers still adopt the Excel file format for a variety of scales statistics, such as the clinical self-care ability scale for Functional Independence Measure. Although people can further analyze Excel files using other statistical analysis software, such as SAS, SPSS, and STATA, they cannot effectively share the archived data in Excel among divisions. We propose to do format conversion on these data and store them in a database. As the collection of Excel files cannot be shared with ease, we plan to use HBase, a non-relational database, to further integrate data. The purpose of this paper is to construct complete import tools and solutions based on HBase to facilitate easy access of data in HBase. Besides, a visual interface is also used to manage HBase to implement user friendly client connection tools for the HBase database.

In this work, we particularly focus on the use of free open-source software, so that end users do not need to spend a huge amount of software license fees. For cloud computing, virtualization technology delivers numerous benefits in addition to being one of the basic roles to build a cloud environment. By virtualization, enterprises can maximize working efficiency without the need to install more facilities in the computer room. In this study, we implemented a virtualization environment and performed experiments on it. The main subject of it is how to use the Open Stack open-source software to build a cloud infrastructure with high availability and a dynamic resource allocation mechanism. It provides a private cloud solution for business and organizations. It belongs to Infrastructure as a Service (IaaS), one of the three service models in the cloud. For the part of the user interface, a web interface was used to reduce the complexity of access to cloud resources for users. We measured the performance of live migration of virtual machines with different specifications and analyzed the data. Also according to live migration modes, we wrote an algorithm to solve the traditional migration problem that needs manually determining whether the machine load is too heavy or not, as a result, the virtual machine load level is automatically detected, and the purpose of automatic dynamic migration to balance resources of servers is achieved.

Along with the improvement of sanitary conditions and changes of life style, people begin to pay attention to the modern concept of health promotion and preventive medicine. Therefore, we built an intelligent environment monitoring feedback system to collect data of physical conditions of employees and air conditions of the working environment, displayed the collected data with a real time interface, and sent out warming messages to prevent accidents. We hope that based on these real-time data, the proposed system can help people make right and timely decisions, and act on time to maintain a beneficial environment in the monitored area.

ABSTRACT In recent years, cloud computing has become popular and its applications widespread. Thus, there exists a common concern, that is, how to arrange and monitor various resources in the cloud computing environment. In the literature, Ganglia and Network Weather Service (NWS) were used to monitor and gather node status and network‐related data, respectively. With supports of Ganglia and NWS, one can effectively administer available resources in the cloud computing environment. In order to achieve high performance of cloud computing, comprehensive monitoring and efficient management are critical. Ganglia is often used to gather status data of resources, such as live states of hosts, CPU or memory utilizations, and surely Ganglia is also capable of monitoring network‐related information; however, instead of Ganglia, we used NWS services to gather network‐related information such as end‐to‐end transmission control protocol/Internet protocol performance data. Compared with Ganglia, NWS services offer more selections and flexibility for measurement schemes. Besides, NWS services could be deployed with nonintruding manner that makes it easier and faster in deploying services to cloud nodes. The network‐related information is acquired immediately after deployment. Although NWS services also provide measurements for CPU and memory utilizations, but less functionality is provided by them than Ganglia in these aspects. Therefore, we combine advantageous features of Ganglia and NWS to achieve the aims of effective monitoring and management of available resources in the cloud environment. Nevertheless, Ganglia and NWS services may not provide sufficient data in realistic situations due to diversified needs of users, especially application developers. For instance, users are not able to directly access utilizations or allocations of resources in the cloud environment via interfaces or channels of Ganglia or NWS. In addition, NWS services based on a domain‐based network information model could greatly decrease overheads caused by unnecessary measurements. Hence, we propose a heuristic QoS measurement approach based on the domain‐based information model. This measurement approach is capable of providing essential information to satisfy user requirements, and thus let users manage and monitor various resources in the cloud environment in a more efficient way. © 2013 Wiley Periodicals, Inc.

With the ever-increasing popularity of the cloud platform, cloud backup scheme attracts more attention from both industry and academia. For cloud providers, profit evaluation of server farms is an important issue of cloud computing economics. Optimal logistic policy should be considered to be an indispensible design simultaneously for providing qualified service to cloud users while the whole cloud center is under construction. To maintain contract-based service quality, the administrator of server farm should adopt the necessary redundancy strategy to provide backup servers when some servers fail. To this aim, focusing upon exploring the optimal profit, two decision parameters of spares and repairmen in the system are considered to maintain regulated service quality for the cloud users. The basic point of our approach is that a novel design approach is developed for evaluating the profit patterns using the finite-source queuing theory. A comprehensive mathematical analysis on profit pattern has been made in detail. Numerical simulation has also been conducted to validate the proposed optimization model. The design illustration is presented to demonstrate engineering application scenario in cloud computing environment, hence the proposed approach indeed provides a feasibly cost-effective design framework to meet logistic economy.

In this paper, we use PCI pass-through technology and make the virtual machines in a virtual environment are able to use the NVIDIA graphics card, which uses the CUDA parallel programming. It makes the virtual machine have not only the virtual CPU but also the real GPU for computing. The performance of virtual machine is predicted to increase dramatically. This paper will measure the performance differences between virtual machines and physical machines by using CUDA; and how virtual machines would verify CPU numbers under influence of CUDA performance. At last, we compare two open source virtualization environment hypervisor, whether it is after PCI pass-through CUDA performance differences or not. Through the experiment, we will be able to know which environment will reach the best efficiency in a virtual environment by using CUDA.

Virtualization is a common strategy to improve the existing computing resources, particularly in cloud computing field. Hadoop, one of Apache projects, is designed to scale up from single servers to thousands of machines, and each offer local computation and storage. However, how to guarantee stability and reliability have become great study topics. In this article, we use current open-source based on software and platform to reach our goal. For instance, Xen-Hyper visor virtualization technology, Open Nebula virtual machines management tool, and so on. After extending component capabilities, we developed a mechanism to support our idea and reached Hadoop High Availability which called Virtualization Fault Tolerance (VFT). We consider a practical problem that occurs frequently in our system, and the results in this paper also confirm the downtime time can be shortened if failure occurred. In this case, it is not only for the Hadoop applications, but also extended to more areas of cluster-based systems.

Due to the trend of individualization and adaptation of e-Learning, more and more SCORM-compliant teaching materials are developed by institutes and individuals in different sites. Also, cloud computing environments are emerging as powerful infrastructures to support e-Learning applications. Therefore, how to rapidly retrieve SCORM-compliant documents on cloud computing environments has become an important issue. Creating an index from folksonomies has been investigated in previous researches; however, the involved uncertainty has not been addressed. This paper focuses on the fuzzy index creation problem for learning content retrieval. A bottom-up approach to constructing the fuzzy index is proposed. The index creation method has been implemented, and a synthetic learning object repository has been built on a Hadoop cloud platform to evaluate the proposed approach. Experimental results show that this method can increase precision of retrieval.

With the development of electronics in governments and business, the implementation of these services are increasing demand for servers. Continued expansion of servers represents our need for more space, power, air conditioning, network, human resources and other infrastructure. Regardless of how powerful servers now become, we do not make good use of all resources and strive for the waste. In this paper, the Green Power Management (GPM) is proposed for load balancing for virtual machine management on cloud. It includes three main phrases: (1) supporting green power mechanism, (2) implementing virtual machine resource monitor onto Open Nebula with web-based interface, and (3) integrating a Dynamic Resource Allocation (DRA) and Open Nebula functions as bases instead of traditionally booting physical machines with command mode.

Today, the multi-core processor has occupied more and more market shares, and the programming personnel also must face the collision brought by the revolution of multi-core processor. Semiconductor scaling limits and associated power and thermal challenges limit performance growth for single-core microprocessors. This reason leads many microprocessor vendors to turn instead to multi-core chip organizations. So programmer or compiler explicitly parallelize the software is the key for enhance the performance on multi-core chip. At the same time, parallel processing is not only the opportunity but also a challenge. The programmer or compiler explicitly parallelize the software is the key for enhance the performance on multi-core chip. In this paper, what we want to know is there any effective way that can reduce our time on rewrite or can automatically parallel the program for multi-processing purpose and do speedup the processing. We discussed some tools that can automatically generate OpenMP directives from serial C/C++ codes, and compare them with each other include normal C/C++ code, and run on general computer and embedded system. Also we compared some tools that are specifically designed to extract the most of data parallelism from C and FORTRAN kernels and translate them into NVIDIA CUDA or OpenCL to know how mush fast after use them.

Abstract Cluster computing is an attractive approach to provide high‐performance computing for solving large‐scale applications. Owing to the advances in processor and networking technology, expanding clusters have resulted in the system heterogeneity; thus, it is crucial to dispatch jobs to heterogeneous computing resources for better resource utilization. In this paper, we propose a new job allocation system for heterogeneous multi‐cluster environments named the Adaptive Job Allocation Strategy (AJAS), in which a self‐scheduling scheme is applied in the scheduler to dispatch jobs to the most appropriate computing resources. Our strategy focuses on increasing resource utility by dispatching jobs to computing nodes with similar performance capacities. By doing so, execution times among all nodes can be equalized. The experimental results show that AJAS can improve the system performance. Copyright © 2011 John Wiley & Sons, Ltd.

Abstract Grid computing employs heterogeneous resources which may be installed on different platforms, hardware/software, computer architectures, and perhaps using different computer languages to solve large‐scale computational problems. As many more Grids are being developed worldwide, the number of multi‐institutional collaborations is growing rapidly. However, to realize Grid computing's full potential, it is expected that Grid participants must be able to share one another's resources. This paper presents a resource broker that employs the multi‐site resource allocation (MSRA) strategy and the dynamic domain‐based network information model that we propose to allocate Grid resources to submitted jobs, where the Grid resources may be dispersed at different sites, and owned and governed by different organizations or institutes. The jobs and resources may also belong to different clusters/sites. Resource statuses collected by the Ganglia, and network bandwidths gathered by the Network Weather Service, are both considered in the proposed scheduling approach. A dynamic domain‐based model for network information measurement is also proposed to choose the most appropriate resources that meet the jobs' execution requirements. Experimental results show that MSRA outperformed the other tested strategies. Copyright © 2010 John Wiley & Sons, Ltd.

Abstract Parallel loop self‐scheduling on parallel and distributed systems has been a critical problem and it is becoming more difficult to deal with in the emerging heterogeneous cluster computing environments. In the past, some self‐scheduling schemes have been proposed as applicable to heterogeneous cluster computing environments. In recent years, multicore computers have been widely included in cluster systems. However, previous researches into parallel loop self‐scheduling did not consider certain aspects of multicore computers; for example, it is more appropriate for shared‐memory multiprocessors to adopt Open Multi‐Processing (OpenMP) for parallel programming. In this paper, we propose a performance‐based approach using hybrid OpenMP and MPI parallel programming, which partition loop iterations according to the performance weighting of multicore nodes in a cluster. Because iterations assigned to one MPI process are processed in parallel by OpenMP threads run by the processor cores in the same computational node, the number of loop iterations allocated to one computational node at each scheduling step depends on the number of processor cores in that node. Experimental results show that the proposed approach performs better than previous schemes. Copyright © 2010 John Wiley & Sons, Ltd.

Nowadays, NVIDIA's CUDA is a general purpose scalable parallel programming model for writing highly parallel applications. It provides several key abstractions - a hierarchy of thread blocks, shared memory, and barrier synchronization. This model has proven quite successful at programming multithreaded many core GPUs and scales transparently to hundreds of cores: scientists throughout industry and academia are already using CUDA to achieve dramatic speedups on production and research codes. In this paper, we propose a hybrid parallel programming approach using hybrid CUDA and MPI programming, which partition loop iterations according to the number of C1060 GPU nodes in a GPU cluster which consists of one C1060 and one S1070. Loop iterations assigned to one MPI process are processed in parallel by CUDA run by the processor cores in the same computational node.

Formerly computer application development just develops in one computer with one application, if one user wants to use some application in other computer (e.g. a computer or device) then the computer or device must be deployed or reinstalled the application in another computer. As a result, due to improve this problem, many application developer start redesign or translate elder application as Web application, early phases technology we commonly see such as CGI, the near future we commonly see such as Java Server Page (JSP), Active Server Page.Net (ASP.net) and PHP Hypertext Page (PHP). At this research we used Java Server Page (JSP) Servlet Technology to translate an old application called as “Cyber Transformer” (CT), JSP is based on Java Enterprise Technology (J2EE) which can also support Java Commodity Grid (CoG) component, and we also translate CT's speeding download algorithm to speed up File Downloading in Grid Environment. This new web application we called it as a new name MIFAS.

In this paper, we use the Hadoop system to build the cloud computing environment. By using data hiding technology to embed data into cover images, we show that the approach using cloud computing would take less execution time than that using a single computer when processing a huge amount of data. Thus, cloud computing provides a convenient platform and also reduces the cost of the equipment required for processing huge amounts of data.

Abstract Data Grids enable the sharing, selection, and connection of a wide variety of geographically distributed computational and storage resources for addressing large‐scale data‐intensive scientific application needs in, for instance, high‐energy physics, bioinformatics, and virtual astrophysical observatories. Data sets are replicated in Data Grids and distributed among multiple sites. Unfortunately, data sets of interest sometimes are significantly large in size, and may cause access efficiency overhead. A co‐allocation architecture was developed in order to enable parallel downloading of data sets from multiple servers. Several co‐allocation strategies have been coupled and used to exploit download rate by specifying among various client–server divides files into multiple blocks of equal sizes to link and address dynamic rate fluctuations. However, one major obstacle, the idle time of faster servers having to wait for the slowest server to deliver the final block, makes it important to reduce differences in finishing time among replica servers. In this paper, we propose a dynamic co‐allocation method, calledRecursively Adjusting Co‐Allocation Method(RACAM), to improve the performance of parallel data file transfer. Our approach reduces the idle time spent waiting for the slowest server and decreases data transfer completion time. We also provide an effective scheme for reducing the cost of reassembling data blocks. Copyright © 2010 John Wiley & Sons, Ltd.

Abstract Data Grids enable the sharing, selection, and connection of a wide variety of geographically distributed computational and storage resources for content needed by large‐scale data‐intensive applications such as high‐energy physics, bioinformatics, and virtual astrophysical observatories. In Data Grids, co‐allocation architectures were developed to enable parallel downloads of data sets from selected replica servers. As Internet is usually the underlying network of a grid, network bandwidth plays as the main factor affecting file transfers between clients and servers. In this paradigm, there are still some challenges that need to be solved, such as to reduce differences in finish times between selected replica servers, to avoid traffic congestion resulting from transferring the same blocks in different links among servers and clients, and to manage network performance variations among parallel transfers. In this paper, we propose theAnticipative Recursively Adjusting Mechanism(ARAM) scheme to adjust the workloads on selected replica servers and handle unpredictable variations in network performance by those servers. Our algorithm is based on using the finish rates for previously assigned transfers to anticipate the bandwidth status for the next section to adjust workloads, and to reduce file transfer times in grid environments. Our approach is useful in grid environments with unstable network link. It not only reduces idle time wasted waiting for the slowest server, but also decreases file transfer completion times. Copyright © 2010 John Wiley & Sons, Ltd.

Loop scheduling on parallel and distributed systems has been a critical problem. Furthermore, it becomes more difficult to deal with on the emerging heterogeneous grid environments. In the past, some loop self-scheduling schemes have been proposed to be applicable to heterogeneous gird environments. In this paper, we propose a performance-based approach, which partitions loop iterations according to the performance weight of nodes. To verify the proposed approach, a grid testbed that consists four schools is built, and matrix multiplication example is implemented to be executed in this testbed. Experimental results show that the proposed approach performs better than previous schemes.

Abstract Over the past few years, research and development in bioinformatics (e.g. genomic sequence alignment) has grown with each passing day fueling continuing demands for vast computing power to support better performance. This trend usually requires solutions involving parallel computing techniques because cluster computing technology reduces execution times and increases genomic sequence alignment efficiency. One example, mpiBLAST is a parallel version of NCBI BLAST that combines NCBI BLAST with message passing interface (MPI) standards. However, as most laboratories cannot build up powerful cluster computing environments, Grid computing framework concepts have been designed to meet the need. Grid computing environments coordinate the resources of distributed virtual organizations and satisfy the various computational demands of bioinformatics applications. In this paper, we report on designing and implementing a BioGrid framework, called G‐BLAST, that performs genomic sequence alignments using Grid computing environments and accessible mpiBLAST applications. G‐BLAST is also suitable for cluster computing environments with a server node and several client nodes. G‐BLAST is able to select the most appropriate work nodes, dynamically fragment genomic databases, and self‐adjust according to performance data. To enhance G‐BLAST capability and usability, we also employ a WSRF Grid Service Portal and a Grid Service GUI desk application for general users to submit jobs and host administrators to maintain work nodes. Copyright © 2008 John Wiley & Sons, Ltd.

We have encountered two challenges when using the PACS system. First, PACS users are limited to certain bandwidths and locations. Second, Web PACS machine replacement is too costly, management is difficult, and better image stability is needed. There are also speed variations for different users at different locations. For instance, radiologists use medical image workstations with direct access to the PACS information system and so have a greater speed for querying and file retrieval. Physicians, on the other hand, use web browsers with no direct access to the PACS information system, which leads to slower network speeds. Physicians also affect one another in overall network speed by processing queries and file retrievals via web browser. There are also insufficient network bandwidth concerns. These often arise when exchanging medical images with other hospitals or downloading large numbers of images. Since these large file volumes are transferred via WANs, insufficient network bandwidths limit upload and download speeds. And if the Web PACS breaks down, the hospital must ask professional engineers for replacements, and spend large amounts of money about a NT$ million or more per unit. This is not only troublesome for system managers, but also costly for the hospital.

The objective of this paper is to implement and evaluate a cluster computing environment by clustering idle PCs (personal computer) with Diskless slave nodes on campuses in order to obtain the effectiveness of the largest computer potency. Two sets of cluster platforms BCCD and DRBL are used to compare parallel computing performance. The objective is to prove that DRBL has better performance than BCCD in this experiment. In order to achieve the objective of a platform for Free Software Teaching, DRBL is applied to the computer classroom, enabling PCs to be manually or automatically switched among different OS (operating system) of Windows, Free Software Teaching and PC Cluster. The bioinformatics program, mpiBLAST, is executed smoothly in the Cluster architecture as well. Through comparing various aspects of performance, including performance of Switch, Swap, this paper is attempted to find out the best Cluster environment in computer classroom at school. Finally, HPCC is used to demonstrate Cluster performance.

In this paper, we propose a resource broker, which providing a friendly interface for accessing available and appropriate resources via user credentials, is developed on a platform constructed by employing the Globus toolkit. This broker not only deploys a domain-based network information model and its dynamic version to measure network status by invoking Network Weather Service (NWS) on grid computing environments, but also uses the Ganglia and NWS tools to monitor resource status and network-related information. Based on the actual value of network bandwidth and preserving the advantages of our previous model, a network bandwidth-award job scheduling algorithm for grid resource brokerage is then developed for communication-intensive job execution. Furthermore, the network information and resource status collected are all up-to-date so that the resource broker can effectively match appropriate grid resources and users’ requests to improve execution.

Effective data distribution techniques can significantly reduce the total execution time of a program on grid computing environments, especially for data mining applications. In this paper, we describe a linear programming formulation for the data distribution problem on grids. Furthermore, a heuristic method, named HDDS (heuristic data distribution scheme), is proposed to solve this problem. We implement the parallel association rule mining method and conduct the experimentations on our grid testbed. Experimental results showed that data mining programs using our HDDS to distribute data could execute more efficiently than traditional schemes could.

In data grid, co-allocation architecture can be used to enable parallel transferring of data file from multiple replicas which stored in the different grid sites. Some schemes base on co-allocation model were proposed and used to exploit the different transfer rates among various client-server network links and to adapt dynamic rate fluctuations by dividing data into fragment. These schemes showed the more fragments used the more performance conducted when data transfer in parallel with evidence. In our previous work, we propose a scheme named anticipative recursively-adjusting mechanism (ARAM) in previous research work. The best thing is performance tuning through the alpha value, it¿s rely on special feature to adapt different network situations in a data grid environment. In this paper, the TCP bandwidth estimation model (TCPBEM) is used to evaluate dynamic link state by detect TCP throughput and packet lost rate between grid nodes. We integrate the model into ARAM, called anticipative recursively-adjusting mechanism plus (ARAM+), that can be more reliable and reasonable then previous one. In the meanwhile, we also design a burst mode which could increase transfer rate of ARAM+. This approach not only adapts worst network link but also speedup the overall performance.

The problem of redundant RFID reader elimination has instigated researchers to propose different optimization heuristics due to the rapid advance of technologies in large scale RFID systems. In this paper, we present a layered elimination optimization (LEO) which is an algorithm independent technique aims to detect maximum amount of redundant readers could be safely removed or turned off with preserving original RFID network coverage. A significant improvement of the LEO scheme is that number of "write-to-tag" operations could be largely reduced during the redundant reader identification phase. Moreover, LEO is a distributed scheme which does not need to collect global information for centralizing control, leading no communications and synchronizations among RFID readers. To evaluate the performance of the proposed techniques, we have implemented the LEO technique along with another redundant reader identification algorithm and other hybrid schemes. In experimental results, the LEO is shown to be effective and provides superior performance in terms of larger number of redundant reader could be detected and with lower algorithm overheads.

Grid computing is increasingly used by organizations to achieve high performance computing and heterogeneous resources sharing. These grids may span several domain administrations via Internet. As a result of this, it may be difficult to monitor, control and manage those machines and resources. This paper aims at providing a multi-platform grid monitoring service which can monitor resources such as CPU speed and utilization, memory usage, disk usage, and network bandwidth in a real-time manner. Monitoring data is extracted form Ganglia and NWS tools then stored and transmitted in XML form and then used for displaying. All the information is displayed using real-time graphs.

In recent years, grid technology, which is frequently used to solve the scientific problem, has ripened gradually. A large number of storage resources and computational power are combined to form a data grid network to deal with massive data of scientific experiments. Data replication generates huge data which are distributed in wide-area for researches around the globe. The files of grid environments can be modified by grid users might bring a critical problem of maintaining data consistency among the several replicas distributed in different machines. For that reason how to maintain the consistency of those files is a great challenge. In this paper, we propose a Oneway Replica Consistency model in data grid environments, which is used for consistency maintenance issue. Furthermore, we anticipate striking a balance between improving data access performance and replica consistency in data grids. This work can find out the more efficient ways of utilizing storage space is an important point.

The sharable content object reference model (SCORM) has become a popular standard for sharing and reusing teaching materials in the field of e-learning. However, efficient retrieval of SCORM-compliant learning objects stored on grid environments is a challenging problem. In this paper, a performance-based approach is proposed to retrieve teaching materials in grid environments. The proposed architecture uses the real-time information gathered by a resource-monitoring tool to estimate the dynamically changing performance of each node, for CPU loading and network bandwidth. In addition, a grid testbed is built to implement the model. Experimental results show that the proposed approach can choose the most appropriate site to reduce the retrieval time

Traditionally, interlibrary cooperation is an important interaction between libraries. Nevertheless, the emerging digital library architecture has not explicitly supported this requirement. In this paper, we proposed an interlibrary cooperation framework for digital libraries using P2P technology. An application of this framework to Faculty Publication Sharing System was presented. Besides, a reputation model based on data mining is utilized to provide libraries with incentives to join this framework

Managing and monitoring a cluster is both a tedious and challenging task, since each computing node is designed as a stand-alone system rather than a part of a parallel architecture. In this paper, a software system that allows the centralized administration of a generic Beowulf cluster is proposed. This system also provides Web services and applications to monitor multiple PC clusters with job submission and scheduling

Grid technology is the great progress of network after Internet, since grid enables the integrated and collaborative use of distributed computing resources owned and managed by multiple organizations, available over a local or wide area network. In this paper, we provided a data grid platform with integrated storage to solve such data grid problems by using PVFS2 to increase the data storage space. The experimental results presented show the effectiveness of such proposed combination of technologies

Dynamic data redistribution is used to enhance data locality and algorithm performance by reducing interprocessor communication in many parallel scientific applications on distributed memory multicomputers. Since the redistribution is performed at runtime, there is a performance tradeoff between the efficiency of the new data decomposition for a subsequent phase of an algorithm and the cost of redistributing data among processors. In this paper, we present a processor replacement scheme to minimize the cost of interprocessor data exchange during runtime. The main idea of the proposed technique is to develop a replacement function for reordering logical processors in the destination phase. Based on the replacement function, a realigned sequence of destination processors can be derived and is then used to perform data decomposition in the receiving phase. Together with local matrix and compressed CRS vectors transposition schemes, the interprocessor communication can be eliminated during runtime. A significant improvement of this approach is that the realignment of data can be performed without interprocessor communication for special cases. The second contribution of the present technique is that the complicated communication sets generation could be simplified by applying local matrix transposition. Consequently, the indexing cost could be reduced significantly. The proposed techniques can be applied in both dense and sparse applications. A generalized symmetric redistribution algorithm is also presented in this work. To analyze the efficiency of the proposed technique, the theoretical analysis proves that up to (p-1)/p data transmission cost can be saved. For general cases, the symmetric redistribution algorithm saves 1/p communication overheads compared with the traditional method. Experimental results also show that the proposed techniques provide superior performance in most data redistribution instances.

This research paper discusses performance issues of cluster and grid computing platforms, and reasons to support the implementation of these computing infrastructures. A number of benchmark programs are executed in these computing systems, in order to perform performance analysis of experimental results. We are able to show that cluster platforms are excellent alternatives to access to supercomputing, due to its cost/performance, scalability and commodity components factors. In addition, we also show that grid technology is viable by increasing total system performance at no additional cost.

In recent years, distributed denial-of-service (DDoS) and denial-of-service (DoS) are the most dreadful network threats. Single-node IDS often suffers from losing its detection effectiveness and capability when processing enormous network traffic. To solve the drawbacks, we propose grid-based IDS, called grid intrusion detection system (GIDS), which uses grid computing resources to detect intrusion packets. For balancing detection load, score subtraction approach (SSA) and score addition approach (SAA) are deployed. Furthermore, to effectively detect intrusions, a two-phase packet detection process is proposed. The first phase detects logical and momentary attacks. Chronic attacks are detected in the second phase. Experiments are also performed and the results show that GIDS is truly an outstanding system in detecting attacks.

In this paper, we present visual tool for performance measurement and analysis of MPI parallel programs in cluster environments. Most of tools available today for cluster systems show solely system performance data (e.g., CPU load, memory usage, network bandwidth, machine-room temperature, server average load, among others), being more suitable for system administrators who maintain such system. The visual tool is designed to show performance data of all computer nodes involved in the execution of MPI parallel program, such as CPU load level and memory usage. Additionally, this tool is able to display comparative performance data charts of multiple executions of the application (instrumented with MPI interface) under development.

The low cost and wide availability of PC-based clusters have made them an excellent alternative to access supercomputing. However, while network of workstations may be readily available, there is an increasing need for performance tools that support these platforms, in order to achieve even higher performance. One of possible ways to increase performance is parallel program restructuring. It is introduced in this paper a toolkit to generate graphical charts for visualization of MPI parallel programs, reflecting to its execution over time, with the use of DP*Graph representation, parallel version of timing graph. In other words, parallel programs are shown through charts its sequential codes, dependencies and communication structures in a particular cluster system platform. Still in this paper, it is discussed the implementation of this toolkit and present some experimental results obtained.

In this paper, a parallel loop self-scheduling scheme for heterogeneous PC cluster systems is proposed. Though the proposed scheme does allow users to choose parameters before the execution initialization phase, there are still weaknesses that motivate us to go further with new improvements in that scheme. For instance, a decision on a fixed and monotonous parameter can easily lead to invalid schedule by using previous input information. Thus, it is proposed in this paper a new scheme, where the scheduling parameter can be adjusted dynamically and fit into most widely available computer systems, in order to provide higher overall performance.

In the present study, THUBioGrid, an experimental distributed computing application for bioinformatics (BioGrid) is proposed. THUBioGrid incorporates directory services (data and software), grid computing methods (security, authentication, data transport and remote jobs), and gene sequence/genomic data processing methods. It uses Java CoG Kit plus bioinformatics Java packages to perform various computational tasks. The performance of THUBioGrid has been tested by executing the FASTA and mpiBLAST programs for protein sequence alignment applications. Results demonstrate the speed-up effects with increasing number of processors used in the computations.

Internet computing and grid technologies promise to change the way we tackle complex problems. They will enable large-scale aggregation and sharing of computational, data and other resources across institutional boundaries. As grid computing is becoming a reality, there is a need for managing and monitoring the available resources worldwide, as well as the need for conveying these resources to the everyday user. This paper describes a resource broker with its main function as to match the available resources to the user's needs. The use of the resource broker provides a uniform interface to access any of the available and appropriate resources using user's credentials. The resource broker runs on top of the Globus toolkit. Therefore, it provides security and current information about the available resources and serves as a link to the diverse systems available in the grid.

In this paper, the authors presented the grid-based decision tree architecture, with the intention of applying it to both parallel and sequential algorithms. Also, it is shown that, based on the scope and model of data mining applied in the grid environment as well as user equivalent perspective, grid roles can be categorized into three types. It was aimed, through these definitions, to help software developers define clear system processes and differentiate the application scope for software applications. To fulfill the architecture, an existing parallel decision tree algorithm was first applied (the SPRINT algorithm) to the grid environment. The performance and differences in many other areas are compared using datasets of different sizes. The experimental results will be used for future reference and further development.

Distributed denial-of-service (DDoS) and denial-of-service (DoS) are the most dreadful network threats in recent years. In this paper, we propose a grid-based IDS, called performance-based grid intrusion detection system (PGIDS), which exploits grid's abundant computing resources to detect enormous intrusion packets and improve the drawbacks of traditional IDSs which suffer from losing their detection effectiveness and capability when processing massive network traffic. For balancing detection load and accelerating the performance of allocating detection node (DN), we use exponential average to predict network traffic and then assign the collected actual traffic to the most suitable DN. In addition, score subtraction algorithm (SSA) and score addition algorithm (SAA) are deployed to update and reflect the current performance of a DN. PGIDS detects not only DoS/DDoS attacks but also logical attacks. Experimental results show that PGIDS is truly an outstanding system in detecting attacks.

In recent years, e-learning has become a popular method of learning. Generally, an e-learning platform which provided multi-media content required a high capacity storage device such as NAS (network attached storage) or SAN (storage area network). However, many schools with insufficient budgets cannot afford this type of expensive equipment. Thus, we employ grid computing and PVFS (parallel virtual file system) technology to integrate the idling storage resources in the school as a means to substitute the purchase of an expensive high-level storage server. In this research we will link several sets of PCs in the school to make them armed with the storage and computing capacity like that of a high-end server. We hope the goal of sharing and reuse of resources among the schools can be achieved.

Traditional loop-scheduling schemes are designed for homogeneous computing environments, and are probably not suitable for emerging heterogeneous PC clusters. This paper applies a two-phased method, named HPLS (Hybrid Parallel Loop Scheduling), to heterogeneous PC-cluster environments. The key idea is to distribute most of the workload to each node according to its performance, which is modeled and estimated in advance. Experiments on our testbed PCcluster showed that, in most cases, our method could reduce the execution time of application programs more than previous schemes could.

In this paper, we introduce our experiments on SLIM and DRBL diskless PC clusters. We constructed them by using 16 machines and only one disk. We run the matrix multiplication and bioinformatics software to evaluate their performance. We find that the performance of SLIM has a great improvement, and the system setup time is much less than before. The network loading of DRBL is also very high. The system boot-up time is no difference between these two systems. The best way to construct the diskless Linux cluster is to eliminate the image size. It can make the nodes have more memory space to handle the program and ease the network loading.

Internet computing and Grid technologies promise to change the way we tackle complex problems. Harnessing these new technologies effectively, it will transform scientific disciplines ranging from highenergy physics to life sciences. This paper describes a metropolitan-scale Grid computing platform named TIGER Project (standing for Taichung Integrating Grid Environment and Resource), which basically interconnects universities and high schools’ cluster computing resources and sharing available resources among them, for investigations in system technologies and high performance applications. This novel project shows the viability of implementation of such project in a metropolitan city.

A cluster is a collection of independent and cheap machines, used together as a supercomputer to provide a solution. A PC cluster consisting of one master node and nine disk-less slave nodes (10 processors), is proposed and built for parallel rendering purposes. The system architecture and benchmark performances of this cluster are also presented. Internet computing and grid technologies promise to change the way we tackle complex problems. They will enable large-scale aggregation and sharing of computational, data and other resources across institutional boundaries. Harnessing these new technologies effectively will transform scientific disciplines ranging from high-energy physics to the life sciences. Also, We construct two heterogeneous PC clusters for parallel rendering purpose and install Linux Red Hat 9 on each PC cluster. Then, these clusters are set to the different subnet. Therefore, we use the grid middleware /spl lambda/obus ToolKit, to connect these two clusters to form a grid computing environment on multiple Linux PC clusters. We also install the SUN Grid Engine, to manage and monitor incoming or outgoing computing jobs and schedule the job to achieve high performance computing and high CPU utilization. The system architecture and benchmark performances of this cluster are also presented.

We integrate the technologies of high availability and load balancing clusters that combine features of both of these cluster types, increasing both the availability and scalability of services and resources. This type of cluster setup is commonly used for Web-based VOD servers.

Abstract The main function of parallelizing compilers is to analyze sequential programs, in particular the loop structure, to detect hidden parallelism and automatically restructure sequential programs into parallel subtasks that are executed on a multiprocessor. This article describes the design and implementation of an efficient parallelizing compiler to parallelize loops and achieve high speedup rates on multiprocessor systems. It is well known that the execution efficiency of a loop can be enhanced if the loop is executed in parallel or partially parallel, such as in a DOALL or DOACROSS loop. This article also reviews a practical parallel loop detector (PPD) that is implemented in our PFPC on finding the parallelism in loops. The PPD can extract the potential DOALL and DOACROSS loops in a program by verifying array subscripts. In addition, a new model by using knowledge‐based approach is proposed to exploit more loop parallelisms in this paper. The knowledge‐based approach integrates existing loop transformations and loop scheduling algorithms to make good use of their ability to extract loop parallelisms. Two rule‐based systems, called the KPLT and IPLS, are then developed using repertory grid analysis and attribute‐ordering tables respectively, to construct the knowledge bases. These systems can choose an appropriate transform and loop schedule, and then apply the resulting methods to perform loop parallelization and obtain a high speedup rate. For example, the IPLS system can choose an appropriate loop schedule for running on multiprocessor systems. Finally, a runtime technique based on the inspector/executor scheme is proposed in this article for finding available parallelism on loops. Our inspector can determine the wavefronts of a loop with any complex indirected array‐indexing pattern by building a DEF‐USE table. The inspector is fully parallel without any synchronization. Experimental results show that the new method can resolve any complex data dependence patterns where no previous research can. One of the ultimate goals is to construct a high‐performance and portable FORTRAN parallelizing compiler on shared‐memory multiprocessors. We believe that our research may provide more insight into the development of a high‐performance parallelizing compiler. Copyright © 2001 John Wiley & Sons, Ltd.

We propose a knowledge based approach for solving loop scheduling problems. A rule based system, called the IPLS, is developed by repertory grid and attribute ordering table to construct the knowledge base. The IPLS chooses an appropriate scheduling algorithm by inferring some features of loops and assigns parallel loops on multiprocessors for achieving high speedup. In addition, the refined system of IPLS can automatically adjust the attributes in a knowledge base according to profile information; therefore IPLS has feedback learning ability.

In this paper, a run-time technique based on inspector-executor scheme is proposed to find available parallelism on loops in this paper. Our inspector can determine the wavefronts by building a DEF-USE table. Additionally, the process of inspector for finding the wavefronts, can be parallelized fully without any synchronization. Our executor can perform the loop iterations concurrently. For each wavefront in a loop, the auto-adapted function is used to get a tailored thread number rather than using fixed thread number for execution. Experimental results show that our new parallel inspector can handle complex data dependency patterns and reduce itself execution time obviously. Besides, the new partitioning strategy for executor can also improve the performance of run-time parallelization obviously.

It is well known that extracting parallel loops plays a significant role in designing parallelizing compilers. The execution efficiency of a loop is enhanced when the loop can be executed in parallel or partial parallel, like a DOALL or DOACROSS loop. This paper reports on the practical parallelism detector (PPD) that is implemented in PFPC (a portable FORTRAN parallelizing compiler running on OSF/1) at NCTU to concentrate on finding the parallelism available in loops. The PPD can extract the potential DOALL and DOACROSS loops in a program by invoking a combination of the ZIV test and the I test for verifying array subscripts. Furthermore, if DOACROSS loops are available, an optimization of synchronization statement is made. Experimental results show that PPD is more reliable and accurate than previous approaches.

We have implemented a portable FORTRAN parallelizing compiler with loop partition on our experimental target system, Acer Altos 10000, running OSF/1 operating system. We have defined a minimal set of thread-related functions and data types, called B Threads, that is required to support the execution of this parallelizing compiler. Our compiler is highly modularized so that the porting to other platforms will be very easy, and it can partition parallel loops into multithreaded codes based on several loop partition algorithms. We have also proposed a general model of parallel compilers, which is an extension from previous model and is useful in constructing a parallelizing compiler for a particular language. The experimental results show that the best speedups are 3.75, 3.46, and 3.81 for matrix multiplication, adjoint convolution, and increasing workload sample, respectively, when the number of processors is four. It has been shown that this approach works and the experimental results are satisfied.