Resource Contention Research Articles

Differences in attitude toward COVID‐19 based on internet and social media usage among community‐dwelling older adults during the first state of emergency

In order to understand the digital divide among older adults during the coronavirus disease 2019 (COVID-19) pandemic, we investigated the association between internet use and compliance with COVID-19 preventive behaviors during the first state of emergency in Japan. A total of 8952 community-dwelling citizens aged 75 years and above were asked about their preventive behaviors during the first state of emergency using a paper-based questionnaire. Among them, 51% responded and were divided into internet users and non-users. We used multivariable logistic regression models to estimate the adjusted odds ratios and 95% confidence intervals of internet use for compliance with preventive behaviors. Approximately 40% of the respondents used the internet, and 9.29% used social media to collect COVID-19-related information. Internet usage was independently associated with compliance with using hand sanitizers, avoiding going out, avoiding eating out, avoiding traveling, getting vaccinated, and getting tested for COVID-19; the adjusted odds ratios (95% confidence intervals) were 1.21 (1.05-1.38), 1.19 (1.04-1.37), 1.20 (1.05-1.38), 1.32 (1.15-1.52), 1.30 (1.11-1.53), and 1.23 (1.07-1.41), respectively. Exploratory subgroup analyses demonstrated that social media users might have shown early adaptation to newly recommended preventive behaviors during the first state of emergency. Results suggest that a digital divide exists, as evidenced by the varied compliance with preventive behaviors depending on internet use. Additionally, social media use may be associated with early adaptation to newly recommended preventive behaviors. Therefore, future studies regarding the digital divide among older adults should investigate differences depending on the types and content of internet resources. Geriatr Gerontol Int 2023; 23: 289-296.

Anthropomorphic diagnosis of runtime hidden behaviors in OpenMP multi-threaded applications

Extreme-scale computing involves hundreds of millions of threads with multi-level parallelism running on large-scale hierarchical and heterogeneous hardware. Some OpenMP multi-threaded applications increasingly suffer from runtime hidden behaviors owning to shared resource contention as well as software- and hardware-related problems. Such hidden behaviors can result in failure and inefficiencies and are among the main challenges in system resiliency. To minimize the impact of hidden behaviors, one must quickly and accurately detect and diagnose the hidden behaviors that cause the failures. However, it is difficult to identify hidden behaviors in the dynamic and noisy data collected by OpenMP multi-threaded monitoring infrastructures. This paper presents an anthropomorphic diagnosis framework for hidden behaviors of OpenMP multi-threaded applications. In the framework, we first design injected heartbeat functions for OpenMP multi-threaded applications. Then, we leverage the heartbeat sequences to extract features of hidden behaviors. Finally, we develop a feature learning-based algorithm using heartbeat analysis, namely HSA, to diagnose hidden behaviors. To evaluate our framework, the NAS Parallel NPB benchmark, EPCC OpenMP micro-benchmark suite, and Jacobi benchmark are used to test the performance of our proposed framework. The experimental results demonstrate that our framework successfully identifies 90.3% of the injected hidden behaviors of OpenMP multi-threaded applications while acquiring low overhead.

CoLocateMe: Aggregation-Based, Energy, Performance and Cost Aware VM Placement and Consolidation in Heterogeneous IaaS Clouds

In many production clouds, with the notable exception of Google, aggregation-based VM placement policies are used to provision datacenter resources energy and performance efficiently. However, if VMs with similar workloads are placed onto the same machines, they might suffer from contention, particularly, if they are competing for similar resources. High levels of resource contention may degrade VMs performance, and, therefore, could potentially increase users’ costs and infrastructure's energy consumption. Furthermore, segregation-based methods result in stranded resources and, therefore, less economics. The recent industrial interest in segregating workloads opens new directions for research. In this article, we demonstrate how aggregation and segregation-based VM placement policies lead to variabilities in energy efficiency, workload performance, and users’ costs. We, then, propose various approaches to aggregation-based placement and migration. We investigate through a number of experiments, using Microsoft Azure and Google's workload traces for more than twelve thousand hosts and a million VMs, the impact of placement decisions on energy, performance, and costs. Our extensive simulations and empirical evaluation demonstrate that, for certain workloads, aggregation-based allocation and consolidation is <inline-formula><tex-math notation="LaTeX">$\sim$</tex-math></inline-formula> 9.61% more energy and <inline-formula><tex-math notation="LaTeX">$\sim$</tex-math></inline-formula> 20.0% more performance efficient than segregation-based policies. Moreover, various aggregation metrics, such as runtimes and workload types, offer variations in energy consumption and performance, therefore, users’ costs.

SLITS: Sparsity-Lightened Intelligent Thread Scheduling

A diverse set of scheduling objectives (e.g., resource contention, fairness, priority, etc.) breed a series of objective-specific schedulers for multi-core architectures. Existing designs incorporate thread-to-thread statistics at runtime, and schedule threads based on such an abstraction (we formalize thread-to-thread interaction as the Thread-Interaction Matrix). However, such an abstraction also reveals a consistently-overlooked issue: the Thread-Interaction Matrix (TIM) is highly sparse. Therefore, existing designs can only deliver sub-optimal decisions, since the sparsity issue limits the amount of thread permutations (and its statistics) to be exploited when performing scheduling decisions. We introduce Sparsity-Lightened Intelligent Thread Scheduling (SLITS), a general scheduler design for mitigating the sparsity issue of TIM, with the customizability for different scheduling objectives. SLITS is designed upon the key insight that: the sparsity issue of the TIM can be effectively mitigated via advanced Machine Learning (ML) techniques. SLITS has three components. First, SLITS profiles Thread Interactions for only a small number of thread permutations, and form the TIM using the run-time statistics. Second, SLITS estimates the missing values in the TIM using Factorization Machine (FM), a novel ML technique that can fill in the missing values within a large-scale sparse matrix based on the limited information. Third, SLITS leverages Lazy Reschedule, a general mechanism as the building block for customizing different scheduling policies for different scheduling objectives. We show how SLITS can be (1) customized for different scheduling objectives, including resource contention and fairness; and (2) implemented with only negligible hardware costs. We also discuss how SLITS can be potentially applied to other contexts of thread scheduling. We evaluate two SLITS variants against four state-of-the-art scheduler designs. We highlight that, averaged across 11 benchmarks, SLITS achieves an average speedup of 1.08X over the de facto standard for thread scheduler - the Completely Fair Scheduler, under the 16-core setting for a variety of number of threads (i.e., 32, 64 and 128). Our analysis reveals that the benefits of SLITS are credited to significant improvements of cache utilization. In addition, our experimental results confirm that SLITS is scalable and the benefits are robust when of the number of threads increases. We also perform extensive studies to (1) break down SLITS components to justify the synergy of our design choices, (2) examine the impacts of varying the estimation coverage of FM, (3) justify the necessity of Lazy Reschedule rather than periodic rescheduling, and (4) demonstrate the hardware overheads for SLITS implementations can be marginal (&lt;1% chip area and power).

Features of the content and disclosure technologies of the information resources of the network of educational libraries under the conditions of the state of war

Features of the content and disclosure technologies of the information resources of the network of educational libraries under the conditions of the state of war

Characterization and reliability of internet resources on pulmonary rehabilitation for individuals with chronic lung disease.

Individuals with lung disease commonly use the internet as a source of health information on pulmonary rehabilitation (PR). The objective of this study was to characterize internet resources on PR, and to assess the content, readability, and quality of patient-directed PR resources. The first 200 websites for the search term 'pulmonary rehabilitation resources and exercise' were analyzed on Google, Yahoo, and Bing. Website content was assessed based on 30 key components of PR from the 2013 and 2021 international consensus statements. Website quality was determined using DISCERN, JAMA benchmarks, and Global Quality Scale (GQS). 66 unique PR websites were identified with the two most common categories being scientific resources (39%) and foundation/advocacy organizations (33%). The average reading level of websites was 11 ± 3. PR content varied significantly across websites (mean range 13.4-21.5). Median DISCERN total score and GQS score were 4 (IQR 3-4) and 3.5 (IQR 2-4), respectively, representing moderate-good quality. Foundation/advocacy websites had higher DISCERN and GQS scores compared to other websites. Foundation/advocacy websites had the highest quality and reliability metrics; however, the higher-than-recommended reading levels may compromise patient comprehension and utilization. This study provides critical insight on the current state of online PR health-related information.

Registered nurse mentors' experiences from co-creation in higher education targeting enhancement of mentorship practices in nursing homes: A qualitative study.

To explore registered nurse (RN) mentors' experiences of participating in the co-creation of a digital educational resource intended to enhance mentorship practices of first-year nursing students in clinical placement in nursing homes. An interpretive, descriptive qualitative study design. Data were collected through two focus group interviews with 15 RN mentors (n=15) participating in co-creative workshops. The co-creative process entailed four co-creative workshops conducted over a 17-month period (June 2019 to end of Oct 2020). Focus group interviews were conducted following the second and third workshops (i.e., in Dec 2019 and in Oct 2020) and data were analysed using thematic analysis. The consolidated criteria for reporting qualitative research (COREQ) checklist was used to report the findings. The analysis identified three themes: (1) co-creative reflective dialogues contributed to knowledge development and increased motivation among mentors; (2) the co-creative approach facilitated and validated the nursing academic-practice tripartite partnership; and (3) effectiveness of workshop structure in facilitating collaboration and mitigating power inequities. A co-creative process provides notable opportunities to advocate for mentorship needs and to enhance mentorship practices in nursing homes. Our study adds to the evidence on co-creation in nursing education, providing insights on the co-creative process and methodology. Higher Education Institutions are uniquely positioned to act as a co-creative arena for the academic-practice collaboration and for the development of educational resources within nursing education. Co-creation may facilitate stronger academic-practice partnership that may more effectively impact mentorship practices in nursing homes and health care system effectiveness. The RNs included in the study were involved in the co-creative process as active contributors informing the digital educational resource content and design.

Quantifying Resource Contention of Co-located Workloads with the System-level Entropy

The workload co-location, such as deploying offline analysis workloads with online service workloads on the same node, has become common for modern data centers. Workload co-location deployment improves data center resource utilization significantly. Still, it also introduces resource contention, resulting in the online service’s quality of service fluctuation, which we call performance interference. As the online service is a tail-latency-sensitive workload, the tail-latency metric can reflect the performance interference degree of the co-location workloads at the application level. However, to guide system design and evaluation, quantitatively evaluating the resource contention of the co-located workloads at the system level is also essential. This article proposes a novel metric called System-Level Entropy (SLE). As a system-level metric, SLE can measure quantitatively resource contention of the co-location systems and perform the apples-to-apples comparison between systems. The experimental results show that SLE can accurately reflect the performance interference of workloads and then evaluate the system resource contention. We also demonstrate two case studies of the SLE. We quantify the affinity of different co-location combinations, including three online services and five offline workloads. Furthermore, we evaluate the effects of state-of-the-art isolated mechanisms (the container and the CPU–affinity binding) with these co-location combinations.

Deep reinforcement learning for application scheduling in resource-constrained, multi-tenant serverless computing environments

Serverless computing has sparked a massive interest in both the cloud service providers and their clientele in recent years. This model entails the shift of the entire matter of resource management of user applications to the service provider. In serverless systems, the provider is highly motivated to attain cost efficient usage of their infrastructure, given the granular billing modules involved. However, due to the dynamic and multi-tenant nature of the serverless workloads and systems, achieving efficient resource management while maintaining function performance is a challenging task. Rapid changes in demand levels for applications cause variations in actual resource usage patterns of function instances. This leads to performance variations in co-located functions which compete for similar resources, due to resource contentions. Most existing serverless scheduling works offer heuristic techniques for function scheduling, which are unable to capture the true dynamism in these systems caused by multi-tenancy and varying user request patterns. Further, they rarely consider the often contradicting dual objectives of achieving provider resource efficiency along with application performance. In this article, we propose a novel technique incorporating Deep Reinforcement Learning (DRL) to overcome the aforementioned challenges for function scheduling in a highly dynamic serverless computing environment with heterogeneous computing resources. We train and evaluate our model in a practical setting incorporating Kubeless, an open-source serverless framework, deployed on a 23-node Kubernetes cluster setup. Extensive experiments done on this testbed environment show promising results with improvements of up to 24% and 34% in terms of application response time and resource usage cost respectively, compared to baseline techniques.

Educational resource of new social media: problem statement

The article is devoted to the problem of the implementation of the educational resource of new social media in the educational process of educational organizations. The author conducts a systematic review of modern scientific and methodological literature, identifies the problem that has developed in the modern theory and practice of using new social media in pedagogical activity, and determines its current state in modern Russia. The article presents the results of a study of the attitude of high school students and their teachers to the use of social media for educational purposes, their ideas about the effectiveness of such a means of teaching and upbringing, about its main strengths and possible threats. The main conclusions of the conducted research are the results of the conducted survey, which suggest that the majority of both high school students and young teachers have a positive attitude to the integration of new social media into the educational process of the school.There are also a lot of questions related to the lack of scientific research of educational opportunities of the developing institute of social media, questions about how to use them in an educational environment, what rules and regulations should be followed, what is the content of a new educational resource, what are its risks and threats, what is the algorithm of its actualization and how to combine it with traditional teaching and upbringing technologies.

Service separation assisted DDoS attack mitigation in cloud targets

DDoS (Distributed denial of service) attacks target on the availability of the victim services. DDoS attacks, being resource intensive attacks, create a heavy resource contention. In the state of the art, we found that resource isolation for legitimate users assists in maintaining service availability even in the presence of DDoS attacks. Resource isolation is an important method to minimize performance interference originated by the resource contention. In this paper, we propose resource isolation through service separation at two levels: physical machine level and virtual machine level. We compare four methods of resource segregation for legitimate users at physical machine level, virtual machine level, container level, and page level. The results of our experiments suggest that adopting different separation levels improve in the response time for legitimate user request even in the presence of DDoS attacks. By using different separation levels, we were able to see a considerable increase in the service availability time of target machines. In addition, there are notable improvements in number of requests failures and response time of target and co-located services.

Characterization of online support group resources for patients with dermatologic conditions.

Dermatologic conditions can have significant quality of life effects on patients. The internet is a first-line accessible resource for patients to seek support and community in managing dermatologic diagnoses. The accessibility and content of online support resources for patients with dermatologic conditions is unclear so we sought to characterize these resources. We conducted online searches utilizing incognito Google, Yahoo, and Bing search engines and identified a total of 36 support group resources. 9 links were for single dermatology support groups and 27 links were for databases of support groups for different dermatologic conditions. We tallied number totals and percentages of online support resources and found wide variability of material in terms of the readability of the group websites, as well as content, medium, and hosts of the groups. Furthermore, we observed an imbalance in representation of resources for certain dermatologic conditions as opposed to others, further highlighting the strong need for the creation of easy-to-access support groups for patients across the spectrum of dermatological disease.

Психологические ресурсы работников организаций для детей-сирот с разным уровнем эмоционального интеллекта

&lt;p&gt;The results of the study of psychological resources of educators of organizations for orphans with different levels of emotional intelligence (EI) are presented. The study involved 186 employees from 39 regions of the Russia. 3 groups were identified: high (N=57), medium (N=83) and lowered (N=34) levels of EI. It is established that the degree of severity and content of psychological resources depend on the level of EI. A group with a high level can be considered as a model. Representatives of the average and lowered level of EI have adaptation mechanisms that compensate for the lack of EI. In a medium group excessively high requirements for their involvement in activities increase the risk of professional burnout, reduce satisfaction with instrumental resources, and increase vulnerability "to a conflict of roles". The specificity of psychological resources is noted, which is expressed in: 1) the predominance of interpersonal EI (hereinafter MEI) over intrapersonal EI (hereinafter WEI); 2) a certain specificity of the structure of psychological resources, which included 5 components: a) the EI resource as the key; b) the resources of stability and self-regulation in relation to subjective well-being; c) motivational resources associated with the potential of compassion; d) environmental resources and the self-efficacy of emotional self-regulation; e) instrumental resources negatively associated with the conflict of roles.&lt;/p&gt;

Queueing System with Two Phases of Service and Service Rate Degradation

In the paper, a queueing system with an unlimited number of servers and two phases of service with degradation in the service rate is studied. The problem of service rate degradation emerges in cloud nodes, where there is contention for hardware resources including computational resources such as CPU cores. In a node, we have a limited number of CPU cores that should execute potentially an unlimited number of processes (requests) in parallel. In our model, the term “server” means a process allocated in the node for execution. So, the number of “servers” is unlimited but their individual performances decrease because CPUs should switch between them during the execution. We consider processes executed in the node with two phases of life cycle that reflects periods with different activity of a process; e.g., in the first phase, the process may require intensive usage of CPU cores but low usage in the second one. Our model distinguishes the phases using different service parameters for them as well as different influence on the service rate degradation in the node. In the paper, two analytical methods are proposed: exact solving of the system of the local balance equation and the asymptotic analysis of the global balance equations. Formulas for the stationary probability distribution of the number of customers in the phases are obtained for both cases. Several numerical examples are provided that illustrate some properties and applicability of the obtained results.

The Effect of Economic Growth, Fiscal Decentralization, Fiscal Stress, and Economic Openness on Regional Inequality

Regional inequality is a phenomenon that occurs universally in all countries, regardless of their size and level of development. Regional inequality is basically caused by differences in the content of natural resources and demographic conditions in each region This study aims to determine the effect of economic growth, the degree of fiscal decentralization, fiscal stress, and economic openness on regional inequality between provinces in Sumatra during the 2010-2020 period. The method used in this study is the panel data regression analysis method using the Random Effect Model. The results showed that simultaneously economic growth, degree of fiscal decentralization, fiscal stress and economic openness influenced regional inequality. Partially, economic growth and the degree of fiscal decentralization have a negative but not significant effect on regional inequality, while fiscal stress has a positive but not significant effect on regional inequality, and economic openness has a positive and significant effect on regional inequality. The implication of this research is that local governments need to adopt policies to impose restrictions on exports and imports, such as barriers in the form of quotas or tariffs.

Athletes’ perceptions of unsupportive parental behaviours in competitive female youth golf

Research exploring athletes and coaches’ perceptions of parent involvement in youth sport has demonstrated the presence of maladaptive parent involvement in youth sport. This research has provided insights into inappropriate parenting practices displayed in the youth sport environment. However, representative of the broader sport parenting literature, data has been primarily gleaned from sports such as tennis and football. In a bid to diversify participant populations, the present study sought to examine female youth golfers’ views of unsupportive parental behaviors within the competitive youth golf environment. Fourteen online synchronous focus groups were conducted with 61 female youth golfers in the specializing (n = 27) and investment (n = 34) stages of development, recruited from seven countries across three continents (Australia, Canada, England, Finland, Ireland, New Zealand, and Scotland). Reflexive thematic analysis revealed three higher-order categories of unsupportive parental behaviors: Emotional Ill-Treatment; Physical Ill-Treatment; and Pressurizing Behavior. Parenting practices consistent with emotional ill-treatment (e.g., verbal ill-treatment) have been previously discussed within the sport parenting literature, however the presence of physical ill-treatment displayed by parents toward child-athletes has not been commonly reported. The present research provides novel insights into athletes’ perceptions of these maladaptive behaviors, and demonstrates that, similar to other youth sports, unsupportive parental behaviors are evident in the youth golf environment. The findings of the current research further reinforce the need for reporting mechanisms for allegations of abuse within youth sport, continued stakeholder education, the development of safeguarding cultures, and also the need to explore parents’ experiences of exhibiting maladaptive involvement. Lay summary: Female youth golfers’ views of unsupportive parental behaviors in the specializing and investment stages of development were explored. Three higher order categories of unsupportive parental behaviors in competitive youth golf were identified: Emotional Ill-Treatment; Physical Ill-Treatment; and Pressurizing Behavior. IMPLICATIONS FOR PRACTICE Findings highlight the need for applied practitioners to work with sport organizations in developing and implementing holistic safeguarding cultures, to prevent child maltreatment in youth sport. Findings can be utilized by applied practitioners, governing bodies, and sport organizations to inform the content of future educational resources and programs for parents of youth golfers,

A sustainable and secure load management model for green cloud data centres

The massive upsurge in cloud resource demand and inefficient load management stave off the sustainability of Cloud Data Centres (CDCs) resulting in high energy consumption, resource contention, excessive carbon emission, and security threats. In this context, a novel Sustainable and Secure Load Management (SaS-LM) Model is proposed to enhance the security for users with sustainability for CDCs. The model estimates and reserves the required resources viz., compute, network, and storage and dynamically adjust the load subject to maximum security and sustainability. An evolutionary optimization algorithm named Dual-Phase Black Hole Optimization (DPBHO) is proposed for optimizing a multi-layered feed-forward neural network and allowing the model to estimate resource usage and detect probable congestion. Further, DPBHO is extended to a Multi-objective DPBHO algorithm for a secure and sustainable VM allocation and management to minimize the number of active server machines, carbon emission, and resource wastage for greener CDCs. SaS-LM is implemented and evaluated using benchmark real-world Google Cluster VM traces. The proposed model is compared with state-of-the-arts which reveals its efficacy in terms of reduced carbon emission and energy consumption up to 46.9% and 43.9%, respectively with improved resource utilization up to 16.5%.

A Task Scheduling Optimization Method for Vehicles Serving as Obstacles in Mobile Edge Computing Based IoV Systems

In recent years, as more and more vehicles request service from roadside units (RSU), the vehicle-to-infrastructure (V2I) communication links are under tremendous pressure. This paper first proposes a dynamic dense traffic flow model under the condition of fading channel. Based on this, the reliability is redefined according to the real-time location information of vehicles. The on-board units (OBU) migrate intensive computing tasks to the appropriate RSU to optimize the execution time and calculating cost at the same time. In addition, competitive delay is introduced into the model of execution time, which can describe the channel resource contention and data conflict in dynamic scenes of the internet of vehicles (IoV). Next, the task scheduling for RSU is formulated as a multi-objective optimization problem. In order to solve the problem, a task scheduling algorithm based on a reliability constraint (TSARC) is proposed to select the optimal RSU for task transmission. When compared with the genetic algorithm (GA), there are some improvements of TSARC: first, the quick non-dominated sorting is applied to layer the population and reduce the complexity. Second, the elite strategy is introduced with an excellent nonlinear optimization ability, which ensures the diversity of optimal individuals and provides different preference choices for passengers. Third, the reference point mechanism is introduced to reserve the individuals that are non-dominated and close to reference points. TSARC's Pareto based multi-objective optimization can comprehensively measure the overall state of the system and flexibly schedule system resources. Furthermore, it overcomes the defects of the GA method, such as the determination of the linear weight value, the non-uniformity of dimensions among objectives, and poor robustness. Finally, numerical simulation results based on the British Highway Traffic Flow Data Set show that the TSARC performs better scalability and efficiency than other methods with different numbers of tasks and traffic flow densities, which verifies the previous theoretical derivation.

Novel Approaches for Resource Management Across Edge Servers

Edge computing aims at reducing computation and storage across the cloud and provides service with reduced latency. Edge devices can be mobile devices, routers, cameras, printers or any Internet of Things (IoT) devices that generate vast amounts of data. The processing of these data is done by virtual machines (VMs) present in the edge servers that are located within close proximity of the edge devices. This work proposes two models which predict resource contention at the edge servers, namely, a Dynamic Markov model for Resource Contention Prediction in Edge Cloud (DMRCP) and a Hybrid Cascade of Regression and Markov model for Resource Contention Prediction (CRMRCP). In DMRCP, a history matrix is updated based on the CPU utilization of a Virtual Machine (VM). This history matrix is used to update a transition probability matrix. This matrix is used to predict the future state of the VM. In the CRMRCP approach, the past CPU utilization values of the virtual machines in the edge servers are used for predicting a set of future CPU utilization values using linear regression, polynomial regression, lasso regression and ridge regression. Then, the predicted future CPU utilization values are used by the dynamic and the second-order Markov models to classify the state of the edge servers as overloaded, underloaded or normally loaded. In both the approaches, the VMs that may cause resource contention are predicted and are migrated to other edge servers such that the destination edge server does not get overloaded after the migration. The DMRCP method is compared with the first-order and the second-order Markov models and the number of VM migrations is analysed to evaluate the performance. The number of VM migrations in the CRMRCP method is compared with that in the second-order Markov model. The overall results prove that the number of VM migrations for the DMRCP is 52.9% less compared to the first-order Markov model and 21.1% less when compared to the second-order Markov model. The number of VM migrations in CRMRCP is reduced by 81.8% when ridge regression cascaded with the second-order Markov model is used.

TurboMGNN: Improving Concurrent GNN Training Tasks on GPU With Fine-Grained Kernel Fusion

<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Graph Neural Networks</i> (GNN) have evolved as powerful models for graph representation learning. Many works have been proposed to support GNN training efficiently on GPU. However, these works only focus on a single GNN training task such as operator optimization, task scheduling, and programming model. Concurrent GNN training, which is needed in the applications such as neural network structure search, has not been explored yet. This work aims to improve the training efficiency of the concurrent GNN training tasks on GPU by developing fine-grained methods to fuse the kernels from different tasks. Specifically, we propose a fine-grained <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Sparse Matrix Multiplication</i> (SpMM) based kernel fusion method to eliminate redundant accesses to graph data. In order to increase the fusion opportunity and reduce the synchronization cost, we further propose a novel technique to enable the fusion of the kernels in forward and backward propagation. Finally, in order to reduce the resource contention caused by the increased number of concurrent, heterogeneous GNN training tasks, we propose an adaptive strategy to group the tasks and match their operators according to resource contention. We have conducted extensive experiments, including kernel- and model-level benchmarks. The results show that the proposed methods can achieve up to 2.6X performance speedup.