Thread Groups Research Articles

GPGPU Task Scheduling Technique for Reducing the Performance Deviation of Multiple GPGPU Tasks in RPC-Based GPU Virtualization Environments

In remote procedure call (RPC)-based graphic processing unit (GPU) virtualization environments, GPU tasks requested by multiple-user virtual machines (VMs) are delivered to the VM owning the GPU and are processed in a multi-process form. However, because the thread executing the computing on general GPUs cannot arbitrarily stop the task or trigger context switching, GPU monopoly may be prolonged owing to a long-running general-purpose computing on graphics processing unit (GPGPU) task. Furthermore, when scheduling tasks on the GPU, the time for which each user VM uses the GPU is not considered. Thus, in cloud environments that must provide fair use of computing resources, equal use of GPUs between each user VM cannot be guaranteed. We propose a GPGPU task scheduling scheme based on thread division processing that supports GPU use evenly by multiple VMs that process GPGPU tasks in an RPC-based GPU virtualization environment. Our method divides the threads of the GPGPU task into several groups and controls the execution time of each thread group to prevent a specific GPGPU task from a long time monopolizing the GPU. The efficiency of the proposed technique is verified through an experiment in an environment where multiple VMs simultaneously perform GPGPU tasks.

GRAM

This article presents GRAM (<underline>G</underline>PU-based <underline>R</underline>untime <underline>A</underline>daption for <underline>M</underline>ixed-precision) a framework for the effective use of mixed precision arithmetic for CUDA programs. Our method provides a fine-grain tradeoff between output error and performance. It can create many variants that satisfy different accuracy requirements by assigning different groups of threads to different precision levels adaptively at runtime . To widen the range of applications that can benefit from its approximation, GRAM comes with an optional half-precision approximate math library. Using GRAM, we can trade off precision for any performance improvement of up to 540%, depending on the application and accuracy requirement.

The Two-Level Semi-Synchronous Parallelization Method for the Caustic and Indirect Luminance Calculation in Realistic Rendering

The paper considers an original approach to the semi-synchronous calculation of the luminance of caustic and indirect illumination for the group of methods based on the bidirectional stochastic ray tracing with backward pho-ton maps. The designed parallelization method uses the two-level threads hierarchy. The low level of this thread hierarchy is synchronous calculations of the part of the whole image defined by a randomly generated pixel mask which is applied to the whole image. The top level is semi-synchronous parallelization level that consists groups of the low level threads which of them calculate own part of the whole image in a way similar to asynchronous calculations. As the top level is semi-synchronous it means that when calculating the luminance of the caustic and indirect illumination, the threads of the low level have access to the data accumulated in the backward photon maps of the other parallel threads of the semi-synchronous level. A special algorithm for organizing an access to data of the upper-level threads avoids de-lays associated with data synchronization. The comparison of the developed solution with purely synchronous and asynchronous parallelization methods is presented.

Data Reuse for Accelerated Approximate Warps

Many data-driven applications, including computer vision, machine learning, speech recognition, and medical diagnostics show tolerance to computation error. These applications are often accelerated on GPUs, but the performance improvements require high energy usage. In this article, we present DRAAW, an approximate computing technique capable of accelerating GPGPU applications at a warp level. In GPUs, warps are groups of threads which issued together across multiple cores. The slowest thread dictates the pace of the warp, so DRAAW identifies these bottlenecks and avoids them during approximation. We alleviate computation costs by using an approximate lookup table which tracks recent operations and reuses them to exploit temporal locality within applications. To improve neural network performance, we propose neuron aware approximation, a technique which profiles operations within network layers and automatically configures DRAAW to ensure computations with more impact on the output accuracy are subject to less approximation. We evaluate our design by placing DRAAW within each core of an Nvidia Kepler Architecture Titan. DRAAW improves throughput by up to $2.8\times $ and improves energy-delay product (EDP) by $5.6\times $ for six GPGPU applications while maintaining less than 5% output error. We show neuron aware approximation accelerates the inference of six neutral networks by $2.9\times $ and improves EDP by $6.2\times $ with less than 1% impact on prediction accuracy.

A thread-adaptive sparse approximate inverse preconditioning algorithm on multi-GPUs

In this study, we present an efficient thread-adaptive sparse approximate inverse preconditioning algorithm on multiple GPUs, called GSPAI-Adaptive. For our proposed GSPAI-Adaptive, there are the following novelties: (1) a thread-adaptive allocation strategy is presented for each column of the preconditioner, and (2) a parallel framework of constructing the sparse approximate inverse preconditioner is proposed on multiple GPUs, and (3) each component of the preconditioner is computed in parallel inside a thread group of GPU. Experimental results show that GSPAI-Adaptive is effective, and is advantageous over the popular preconditioning algorithms in two public libraries, and a latest parallel sparse approximate inverse preconditioning algorithm.

Incidence of Postoperative Adhesions after Laparoscopic Myomectomy with Barbed Suture

Aim: The aim of the study was to assess the incidence of postoperative adhesion by baseball running suture using barbed suture (BS) in laparoscopic myomectomy in comparison with sutures using an absorbable thread. Methods: Two hundred fifteen patients who underwent second-look laparoscopy (SLL) 6 months after laparoscopic myomectomy at our hospital between 2010 and 2014 were retrospectively reviewed. The incidence, numbers, types, and extent of adhesions were evaluated according to the more comprehensive adhesion scoring method. Propensity score matching (PS) (1:1) between the groups was performed by using the diameter of the largest myoma, the number of enucleated myomas, and the type of adhesion barrier. Results: Running baseball sutures and running sutures were applied to 28 and 187 patients with unidirectional BS and absorbable thread, respectively, to close the incised serosal wounds enucleating largest myomas during laparoscopic myomectomy. After PS matching for the patients, surgical findings and the incidence of postoperative wound adhesions were compared between the groups, including 22 patients each. The surgical findings were similar between the groups, except for the total surgical duration and blood loss (medians), which were significantly shorter and lower in the BS group than in the absorbable thread group (70 vs. 100 min; p = 0.01, 50 vs. 100 mL; p = 0.02). Regarding findings of SLL, no significant differences in the incidence of postoperative wound adhesions were found between the groups (BS, 4/22 [18.2%] versus absorbable thread, 8/22 [36.4%]; p = 0.31). Conclusion: Our data indicated that the incidence of postoperative adhesion following the use of BS for wound closure in laparoscopic myomectomy was similar to that following the use of conventional suture.

Influence of thread shapes of custommade root-analogue implants on stress distribution of peri-implant bone: A three-dimensional finite element analysis

To explore the effects from the thread shapes of custom-made root-analogue implant (RAI) on distributions of von Mises stress around the peri-implant bone. Five one-stage RAI three-dimensional finite element (FE) models with different thread shapes (V-shaped design, square design, buttress design, reverse buttress design and none thread design) and congruent bone were created through reverse engineering technology. The data of the five models were imported into the FE analysis software to calculate. A force of 100 N was applied parallelly and of 45° to the implant axis respectively. Analysis was performed to evaluate the von Mises stress distributions at the peri-implant regions with the help of the Ansys 16 software. The von Mises stresses distributed mostly at the implant cervical regions and the tip ends of the threads on the cortical bone under oblique loading, while on the cancellous bone, the stresses concentrated mostly on the implant lateral cervical regions, the tip ends of the threads and the apical regions. When under vertical loading, the von Mises stresses distributed mostly at the implant cervical regions on the cortical bone while at the tip ends of the threads and the lateral apical regions on the cancellous bone. The von Mises stresses were better distributed on the thread groups under both kinds of loadings compared with no thread design. But there was no obvious difference among the different thread groups. The concentrations of the von Mises stresses on the cancellous bone in the thread groups were mostly at the tip ends of the threads while less in the apical area. The von Mises stresses were better distributed on the cancellous bone on the other three thread designs than on square design. Thread designs are advocated for the reason that adding thread designs to the RAI standard design will have a positive effect on stress distributions at the peri-implant regions and it will reduce the concentrations of von Mises stresses on the cortical bone. From the standpoint of the stress distribution, V-shaped design, buttress design and reverse buttress design are more suitable for RAI than square design. There is no difference of the distributions of the von Mises stresses in the RAI between different thread designs.

An efficient sparse approximate inverse preconditioning algorithm on GPU

SummaryThe sparse approximate inverse (SPAI) preconditioner has proven to be effective in accelerating the convergence of iterative methods. Recently, accelerating it on the graphics processing unit (GPU) has attracted considerable attention due to the fact that the cost of constructing it is high. This motivates us to investigate how to accelerate the construction of SPAI preconditioners on GPU in this paper. We propose an efficient sparse approximate inverse algorithm on GPU, called SPAI‐Adaptive. For our proposed SPAI‐Adaptive, there are the following novelties: (1) an adaptive thread allocation strategy for SPAI‐Adaptive is proposed to assign the optimal thread number for each column of the preconditioner, and (2) Each component of the preconditioner, which includes finding indices I and J, constructing local submatrix, decomposing the local matrix into QR, and solving the upper triangular linear system, is computed in parallel inside a thread group of GPU. Experimental results show that the proposed SPAI‐Adaptive is effective, and has good performance and high parallelism.

Thread/OpenThread: A Compromise in Low-Power Wireless Multihop Network Architecture for the Internet of Things

Extending an Internet subnet by connecting resource-constrained nodes (e.g., embedded sensors and actuators) over multiple wireless hops is necessary to support the future Internet of Things (IoT). RPL, the IPv6 routing standard for low-power and lossy networks, tried to achieve this goal but has not seen wide adoption in practice. As an alternative, Thread is a recently standardized low-power network protocol for IoT, driven by the Thread group, an industry consortium led by Google/Nest. We provide a comparative analysis of the technical aspects of RPL and Thread based on their specifications, explaining why using Thread, as opposed to RPL, may make sense for the future Internet. Specifically, the fundamental differences between RPL and Thread are their respective scopes and multihop network architectures, which result in Thread's unique design and advantages over RPL. Lastly, we evaluate Thread in an indoor multihop wireless testbed using OpenThread, an official open source implementation of Thread. This work serves as the first analysis of the Thread protocol in academia.

GVTS: Global Virtual Time Fair Scheduling to Support Strict Fairness on Many Cores

Proportional fairness in CPU scheduling has been widely adopted to fairly distribute CPU shares corresponding to their weights. With the emergence of cloud environments, the proportionally fair scheduling has been extended to groups of threads or nested groups to support virtual machines or containers. Such proportional fairness has been supported by popular schedulers, such as Linux Completely Fair Scheduler (CFS) through virtual time scheduling. However, CFS, with a distributed runqueue per CPU, implements the virtual time scheduling locally . Across different queues, the virtual times of threads are not strictly maintained to avoid potential scalability bottlenecks. The uneven fluctuation of CPU shares caused by the limitations of CFS not only violates the fairness support for CPU assignments, but also significantly increases the tail latencies of latency-sensitive applications. To mitigate the limitations of CFS, this paper proposes a global virtual-time fair scheduler (GVTS) , which enforces global virtual time fairness for threads and thread groups, even if they run across many physical cores. The new scheduler employs the hierarchical enforcement of target virtual time to enhance the scalability of schedulers, which is aware of the topology of CPU organization. We implemented GVTS in Linux kernel 4.6.4 with several optimizations to provide global virtual time efficiently. Our experimental results show that GVTS can almost eliminate the fairness violation of CFS for both non-grouped and grouped executions. Furthermore, GVTS can curtail the tail latency when latency-sensitive applications are co-running with batch tasks.

Effect of medicated thread moxibustion on apoptosis of hippocampal neurons in rat models of chronic cerebral ischemic vascular dementia

To investigate the effect of medicated thread moxibustion onapoptosis in hippocampal neuronsin a rat model of chronic cerebral ischemic vascular dementia. A total of 40 male Wistar rats were randomly divided into normal group and sham-operated group (7 rats each), and rat model of chronic cerebral ischemic vascular dementia (14 rats). The model group rats were treated with medicated thread moxibustion two weeks after surgery, once a day, with one day break every six days, (24 times in all)and an observation period of 4 weeks. At the end of therapy, H&E staining was used to monitor changes in the neurons in CA1 area of the rat hippocampus. Changes in related indexes such as Bax, Bc1-2 and C-fos of neuron apoptosis in hippocampus CA1 area were determined by immunohistochemistry, while protein expression was semi-quantitatively assayed using imaging analysis technique. There was significant hippocampal neuronal necrosis six weeks after model establishment, but the necrosis was milder in rats in the medicated thread moxibustion group. Bax and C-fos were positively expressed and significantly higher in the hippocampus of chronic cerebral ischemic vascular dementia rats (model group) than in the medicated thread moxibustion group after treatment (p <0.01). The expression of Bcl-2 was increased in the medicated thread group after treatment, and was higher in the model group, but comparable to that in the sham-operated group (p=0.975>0.05). Medicated thread moxibustion alleviates hippocampal neuronal necrosis, inhibits neuron apoptosis in hippocampus CA1 area, protects nerves, and maintains relative equilibrium inBax/Bcl-2 through down-regulation of C-fos and Bax and up-regulation of Bcl-2. Thus, cell apoptosis-related pathway may be one of its mechanisms of action.

Shared Last-Level Cache Management and Memory Scheduling for GPGPUs with Hybrid Main Memory

Memory intensive workloads become increasingly popular on general purpose graphics processing units (GPGPUs), and impose great challenges on the GPGPU memory subsystem design. On the other hand, with the recent development of non-volatile memory (NVM) technologies, hybrid memory combining both DRAM and NVM achieves high performance, low power, and high density simultaneously, which provides a promising main memory design for GPGPUs. In this article, we explore the shared last-level cache management for GPGPUs with consideration of the underlying hybrid main memory. To improve the overall memory subsystem performance, we exploit the characteristics of both the asymmetric read/write latency of the hybrid main memory architecture, as well as the memory coalescing feature of GPGPUs. In particular, to reduce the average cost of L2 cache misses, we prioritize cache blocks from DRAM or NVM based on observations that operations to NVM part of main memory have a large impact on the system performance. Furthermore, the cache management scheme also integrates the GPU memory coalescing and cache bypassing techniques to improve the overall system performance. To minimize the impact of memory divergence behaviors among simultaneously executed groups of threads, we propose a hybrid main memory and warp aware memory scheduling mechanism for GPGPUs. Experimental results show that in the context of a hybrid main memory system, our proposed L2 cache management policy and memory scheduling mechanism improve performance by 15.69% on average for memory intensive benchmarks, whereas the maximum gain can be up to 29% and achieve an average memory subsystem energy reduction of 21.27%.

Clinical effects of circumcision device combine with elastic thread in posthetomy: analysis of 1362 cases

Objective To observe the clinical effects of the circumcision device combine with elastic thread for redundant prepuce and phimosis in posthetomy. Methods Clinical data of 1362 patients of redundant prepuce and phimosis underwent circumcision device combine with elastic thread were analyzed. The operative time, recovery time and main complication with the Shang Ring have been compared. Results All operations were successfully completed. Circumcision device combine with elastic thread group avoid ring-removal pain, the healing time and wound edema incidence was significantly reduced, as well as the pain after the operation. Conclusions Circumcision device combine with elastic thread has great advantages and wide introduction value. Key words: Circumcision, Male

Immediate mechanical stability of threaded and porous implant systems

BackgroundPrimary stability of a dental implant system is an essential factor to maintain its long-term success. Thus, the objective of this study was to examine whether primary stability is different between threaded and porous dental implant systems placed in artificial bone blocks and human cadaveric mandibular bone. Materials and methodsForty-two threaded and 42 highly porous dental implants were placed in artificial polyurethane bone foams with 7 different thicknesses (3.5 to 12mm). In addition, 11 threaded and 11 porous implants were installed in 8 edentulous mandibles of human cadavers. Implant stability quotient values, insertion torque, static and dynamic stiffness, and viscoelastic tan δ of each implant system were measured. Mean gray values were obtained at the implantation sites in the human mandible. FindingsThe porous implant group had substantially lower implant stability quotient values and insertion torque values than the threaded implant group that were equal or >5.5mm in thickness of the artificial bone block (p<0.026) with the exception of 8.5mm thickness, while static and dynamic stiffness values were not different between the two implant groups greater than 5.5mm in thickness (p>0.132). Static and dynamic stiffness values of the porous group were significantly greater than the thread group in the human mandibular bone (p<0.015). InterpretationThe porous layer supports axial loading better than lateral and shear loading of the dental implant system. This result indicates that trabecular shaped architecture of the porous layer may provide sufficient anchorage compromising reduction of the axial primary stability of the porous implant system to be comparable with the threaded implant system.

A Survey on Secure Communication Protocols for IoT Systems

The Internet of Things (IoT) integrates a large number of physical objects that are uniquely identified, ubiqui- tously interconnected and accessible through the Internet. IoT aims to transform any object in the real-world into a computing device that has sensing, communication and control capabili- ties. There is a growing number of IoT devices and applications and this leads to an increase in the number and complexity of malicious attacks. It is important to protect IoT systems against malicious attacks, especially to prevent attackers from obtaining control over the devices. A large number of security research solutions for IoT have been proposed in the last years, but most of them are not standardized or interoperable. In this paper, we investigate the security capabilities of existing protocols and networking stacks for IoT. We focus on solutions specified by well-known standardization bodies such as IEEE and IETF, and industry alliances, such as NFC Forum, ZigBee Alliance, Thread Group and LoRa Allianc.

Influence of Thread Design on Dental Implant Osseointegration Assayed Using the Lan-Yu Mini-Pig Model

The primary success of implantation is according to osseointegration. The bone-to-implant contact (BIC) % is one of the methods to evaluate the osseointergation. In our study, the mini-pig animal model was used to study the osseointegration in different implant macro-design in early implantation period. Eight mini-pigs were divided into 2 groups and implanted with 2 thread profiles (trapezoidal and square) and 5 thread pitches (0.6, 0.8, 1.0, 1.2, and 1.4 mm) into their jaws. After 8 and 16 weeks, the mini-pigs were sacrificed and the specimen of implant surrounded bone were harvested. The implant surface was detected under scanning electron microscopy (SEM), Masson-Goldner staining (MGS) and immunohistochemical staining (IHC). BIC were obtained under SEM and MGS. Bone formation and maturation were assessed using IHC to detect bone morphogenetic protein-2 (BMP-2) and osteocalcin (OC). The BIC measured for trapezoidal-thread implants with a pitch of 0.8 mm was higher than that of square-thread implants in both the 8- and 16-week groups. The BIC values of the trapezoidal thread profile with a pitch range of 0.6–1.4 mm were not significantly different (P > 0.05) in the 8- and 16-week groups. BMP-2 and OC were detected in both cases, but BMP-2 and OC in the trapezoidal thread group was more regular than the square thread group. The bone in the trapezoidal thread group was more mature. In conclusion, the trapezoidal thread profile with 0.8 mm pitch is the most ideal macro-design for osseointegration.

Thread Data Sharing in Cache

On modern multi-core processors, independent workloads often interfere with each other by competing for shared cache space. However, for multi-threaded workloads, where a single copy of data can be accessed by multiple threads, the threads can cooperatively share cache. Because data sharing consolidates the collective working set of threads, the effective size of shared cache becomes larger than it would have been when data are not shared. This paper presents a new theory of data sharing. It includes (1) a new metric called the shared footprint to mathematically compute the amount of data shared by any group of threads in any size cache, and (2) a linear-time algorithm to measure shared footprint by scanning the memory trace of a multi-threaded program. The paper presents the practical implementation and evaluates the new theory using 14 PARSEC and SPEC OMP benchmarks, including an example use of shared footprint in program optimization.

Gold thread implantation promotes hair growth in human and mice

Thread-embedding therapy has been widely applied for cosmetic purposes such as wrinkle reduction and skin tightening. Particularly, gold thread was reported to support connective tissue regeneration, but, its role in hair biology remains largely unknown due to lack of investigation. When we implanted gold thread and Happy Lift™ in human patient for facial lifting, we unexpectedly found an increase of hair regrowth in spite of no use of hair growth medications. When embedded into the depilated dorsal skin of mice, gold thread or polyglycolic acid (PGA) thread, similarly to 5% minoxidil, significantly increased the number of hair follicles on day 14 after implantation. And, hair re-growth promotion in the gold threadimplanted mice were significantly higher than that in PGA thread group on day 11 after depilation. In particular, the skin tissue of gold thread-implanted mice showed stronger PCNA staining and higher collagen density compared with control mice. These results indicate that gold thread implantation can be an effective way to promote hair re-growth although further confirmatory study is needed for more information on therapeutic mechanisms and long-term safety.

An Observational Study of the Recurring Formation and Dissipation of a Dynamic Filament

Based on observations at the \(\mbox{H}\upalpha\) wavelength from the Hinode spacecraft, we report here the detailed process of a dynamical filament that showed repeated appearance and dissipation in a filament channel. First, \(\mbox{H}\upalpha\) short fibrils spreading in the pre-formed filament channel joined into longer threads. The joining process was found to be accompanied by small-scale brightening activity, indicating the possible involvement of magnetic reconnection. The forming filament was thickened by merging the neighboring dark threads that were nearly parallel to the axis and also those adjacent to its main endpoints. The formed filament as a single coherent structure only existed for tens of minutes, immediately followed by the dissipation. The dissipation appeared to start with expansion of the filament body, ascending and stripping away of the filament threads, and mass drainage along the legs of the filament. The formation–dissipation process of the filament was repeated three times within the four-hour observational window of Hinode. These observations indicate that the filament structure is highly dynamic. This study provides the observational evidence to confirm the hypothesis of Martin et al. (Ann. Geophys.26, 3061, 2008) on the irreversible build-up of magnetic fields in the corona by discrete threads or groups of threads ascending bodily into the corona.

DagTM: An Energy-Efficient Threads Grouping Mapping for Many-Core Systems Based on Data Affinity

Many-core processors are becoming mainstream computing platforms nowadays. How to map the application threads to specific processing cores and exploit the abundant hardware parallelism of a many-core processor efficiently has become a pressing need. This work proposes a data affinity based threads grouping mapping strategy Data Affinity Grouping based Thread Mapping (DagTM), which categorizes threads into different groups according to their data affinity and the hardware architecture feature of many-core processors. After that, the thread groups are mapped to the specific processing cores to be energy efficiently executed. More specifically, first, the intra-thread data locality is analyzed by computing the data reuse distance, and the inter-thread data affinity is quantified by affinity matrix. Second, the threads are categorized into different groups via affinity subtree spanning algorithm. Finally, the thread groups are assigned to different processing cores through static binding. DagTM is able to reduce conflicts of the shared memory access and additional data transmission, increase utilization of the computing resources, and reduce entire system energy consumption. Experimental results show that DagTM obtains a nearly 14% improvement in computing performance, and a nearly 10% reduction in energy consumption compared with the traditional thread mapping mechanism under the condition of not introducing additional runtime overhead.