Workload Model Research Articles

DxPU: Large-scale Disaggregated GPU Pools in the Datacenter

The rapid adoption of AI and convenience offered by cloud services have resulted in the growing demands for GPUs in the cloud. Generally, GPUs are physically attached to host servers as PCIe devices. However, the fixed assembly combination of host servers and GPUs is extremely inefficient in resource utilization, upgrade, and maintenance. Due to these issues, the GPU disaggregation technique has been proposed to decouple GPUs from host servers. It aggregates GPUs into a pool and allocates GPU node(s) according to user demands. However, existing GPU disaggregation systems have flaws in software-hardware compatibility, disaggregation scope, and capacity. In this article, we present a new implementation of datacenter-scale GPU disaggregation, named DxPU. DxPU efficiently solves the above problems and can flexibly allocate as many GPU node(s) as users demand. To understand the performance overhead incurred by DxPU, we build up a performance model for AI specific workloads. With the guidance of modeling results, we develop a prototype system, which has been deployed into the datacenter of a leading cloud provider for a test run. We also conduct detailed experiments to evaluate the performance overhead caused by our system. The results show that the overhead of DxPU is less than 10%, compared with native GPU servers, in most of user scenarios.

The globus compute dataset: An open function-as-a-service dataset from the edge to the cloud

We present a unique function-as-a-service (FaaS) dataset capturing the use of the Globus Compute (previously funcX) platform. Globus Compute implements a federated model via which users may deploy endpoints on arbitrary remote computers, from the edge to high performance computing (HPC) cluster, and they may then invoke Python functions on those endpoints via a reliable cloud-hosted service. The dataset covers 31 weeks and includes 2121472 task submissions from 252 users executed on 580 remote computing endpoints. It includes 277386 registered functions. We describe the dataset and various observations, some that are similar to other FaaS datasets, for example, that 74% of tasks run for less than 1 s, and some that are unique to Globus Compute, for example, that endpoints are used in different ways and that the majority of functions are related to scientific computing and machine learning. To the best of our knowledge, this dataset represents the first federated FaaS dataset that includes user workloads, distributed computing endpoints, and analysis of registered function bodies. We expect the dataset to be useful for researching FaaS architectures, workload modeling, container warming, and other distributed computing architectures.

Applying self-powered sensor and support vector machine in load energy consumption modeling and prediction of relational database

This study explores the analysis and modeling of energy consumption in the context of database workloads, aiming to develop an eco-friendly database management system (DBMS). It leverages vibration energy harvesting systems with self-sustaining wireless vibration sensors (WVSs) in combination with the least square support vector machine algorithm to establish an energy consumption model (ECM) for relational database workloads. Through experiments, the performance of self-sustaining WVS in providing power is validated, and the accuracy of the proposed ECM during the execution of Structured Query Language (SQL) statements is evaluated. The findings demonstrate that this approach can reliably predict the energy consumption of database workloads, with a maximum prediction error rate of 10% during SQL statement execution. Furthermore, the ECM developed for relational databases closely approximates actual energy consumption for query operations, with errors ranging from 1 to 4%. In most cases, the predictions are conservative, falling below the actual values. This finding underscores the high predictive accuracy of the ECM in anticipating relational database workloads and their associated energy consumption. Additionally, this paper delves into prediction accuracy under different types of operations and reveals that ECM excels in single-block read operations, outperforming multi-block read operations. ECM exhibits substantial accuracy in predicting energy consumption for SQL statements in sequential and random read modes, especially in specialized database management system environments, where the error rate for the sequential read model is lower. In comparison to alternative models, the proposed ECM offers superior precision. Furthermore, a noticeable correlation between model error and the volume of data processed by SQL statements is observed. In summary, the relational database ECM introduced in this paper provides accurate predictions of workload and database energy consumption, offering a theoretical foundation and practical guidance for the development of eco-friendly DBMS.

Time, workload model and the entrepreneurial construction of the neoliberal academic

This paper analyses the workload model of a UK business school. The workload model is conceived as a time budget, allocating time units to realise academic tasks. The paper highlights how this tool participates in the government of the academic population, in tandem with an individual performance measurement system. On the one hand, the multiple and competing academic tasks are independently regulated by a series of disciplinary objectives. On the other hand, they all need to be realised within a limited timeframe. Yet the workload model estimates a quantity of time resources deemed necessary to deliver on these objectives. The whole creates an apparatus of security - an ensemble of discourses and technologies that work together to control, manage and shape a population - wherein the workload model enacts competition through time to realise a variety of highly disciplined academic tasks. Thus, the academic must make choices, including the “investment” of his/her free time. Through this process the academic is subjectivised as an entrepreneur of the self.

Mental workload evaluation model of receiver aircraft pilots based on multiple resource theory

AbstractAerial refueling is an extender of air combat capability that has received widespread attention with the development of the military field, particularly the mental workload of the pilot performing the aerial refueling task, as it is the key to the aerial refueling success rate. Therefore, this study analyzes the behavior of receiver aircraft pilots from rendezvous to the separation phase and constructs a mental workload model to improve the aerial refueling success rate. First, the time‐channel‐action unit network was constructed based on Petri net and multiple resource theory (MRT). Second, the mental workload evaluation model was constructed from three dimensions: conventional resource occupation, additional resource occupation, and time occupation. Finally, a simulated experiment was conducted, and the results showed that the mental workload obtained by the model constructed in this study exhibits a high positive correlation with subjective mental workload, pupil diameter, and blink rate, surpassing the accuracy of the traditional MRT model.

Predictive Mental Workload Modeling Methodology for Single-Pilot Operations System Design

Predictive mental workload model is an important tool for evaluating early single-pilot operations (SPO) system design and can be used to explore the impact of different function allocation decisions before prototype or simulator implementation. To build this model, this paper proposes a methodology based on task analysis. Firstly, it selects the approach and landing scenario of the current two-pilot operations (TCO) and completes the task description of the TCO using the hierarchical task analysis. Secondly, it identifies the function allocation requirements for the transition from TCO to SPO using the work domain analysis and proposes the original function allocation decisions based on the SPO’s system framework. Finally, it creates the predictive mental workload model for evaluating different SPO system design options based on temporal network and multiple resource theory. The researchers use the created model to evaluate the original function allocation decisions and find that the designed SPO system has instantaneous workload values that exceed the upper limit at multiple time points. By analyzing the causes of these problems, the researchers introduce audible warnings and visual indicators to the SPO system to improve design decisions and evaluate these decisions. The results show that the continuous workload profile and overall average workload of the redesigned SPO system are better than the current TCO system, and also prove that the predictive mental workload model can be well applied to evaluate the system design of early SPO.

Simulation modeling of Zoom traffic on a campus network: A case study

In this paper, we develop a synthetic workload model for the Zoom network application based on empirical Zoom traffic measurements from a campus network. We then use this model in a simulation study of Zoom network traffic at the campus scale. The simulation results show that hybrid learning places a substantial load on the campus network. Additional simulation experiments investigate the potential benefits of locally-hosted Zoom infrastructure, improved load balancing strategies for Zoom servers, and multicast delivery for Zoom network traffic. The simulation results show that the multicast approach offers the greatest potential benefit for improving Zoom performance on our campus network.

Characterization of I/O Behaviors in Cloud Storage Workloads

As cloud platforms become increasingly popular, accurately understanding I/O behaviors in modern cloud storage is of paramount importance for system design and optimization. This paper sheds new light on the correlation of inter-arrival times of both read and write requests at the block level in four representative cloud storage workloads - AliCloud, Systor'17, MSRC and FIU. Our study reveals that I/O arrivals at the block level are very complex in modern cloud storage. There is a certain degree of correlation in the long-term timescale for request arrival intervals in AliCloud and Systor'17_read. Request arrival intervals in MSRC, FIU and Systor'17_write, however, are almost uncorrelated. The Gaussianity test confirms that I/O burstiness appears to be Gaussian in AliCloud_write and Systor'17_read, but the burstiness is non-Gaussian in other workloads. Importantly, we unfold the existence of self-similarity in cloud storage workloads with a certain degree of correlations, via visual evidence, the autocorrelation structure of the aggregated process of I/O request sequences, and Hurst parameter estimates. We further design an alpha-stable workload model for synthetic I/O generation, and the experimental results demonstrate that our model has an edge over conventional models in terms of accurately emulating I/O burstiness.

Technological innovation for workload allocation in nursing care management: an integrative review.

Technology reduces the nursing workload, improve the quality care processes, patient's safety, and avoid staff burnout. Innovative technologies are disrupting healthcare systems by improving the efficiency of processes and management. There is a discussion on the benefits, challenges, and barriers of these technologies and considering human factors of nursing management. The aim was to analyze the influence of technologies on the distribution of workload for nursing care management. An integrative literature review was performed. Four databases were searched: Scopus, Scielo, PUBMED, and CINALH following PRISMA guidelines. Articles published from January 2016 to December 2020, published in English, Spanish and Portuguese were included. Studies were excluded when they were not original research, did not met the quality criteria or they did not answer the research questions. Quality appraisal was performed using the Crowe Critical Appraisal Tool version 1.4 (CCAT). Two reviewers independently examined the title and abstract for eligibility according to the inclusion and exclusion criteria. 2818 potentially relevant articles were found, but once the inclusion and exclusion criteria in the abstracts were analyzed, 177 remained for evaluation. After following the PRISMA Guidelines, 35 studies were included in the review. Three categories were identified: Nursing workload; Information technologies and technological means for management; Technology acceptance. Technology has the potential to improve care management by estimating nurse workload in ICUs and non-critical units, but scientific evidence is more detailed in the former type of services. The literature provides insights about the factors that factors and the barriers that promote the technology acceptance and usability. We did not find studies comparing technologies and no scientific evidence proving improvements in care.

Time Required for Gross Examination of Routine Second and Third Trimester Singleton Placentas by Pathologists' Assistants.

In both Canada and the United States, workload measurement for anatomic pathology is mainly based on complexity and clinical significance of specimens, with gross examination being a considerable contributor. While Pathologists' Assistants (PAs) play an increasing role in gross examination, there is little known regarding the time required for PAs to complete grossing tasks. This information is essential for effective staffing and workload management in pathology laboratories. The objective of our study was to determine the time required for PAs to gross second and third trimester singleton placentas in a large tertiary hospital with a significant perinatal pathology service. For our study, 7 certified PAs each grossed a minimum of 10 second and third trimester singleton placentas using a standard placental grossing protocol, an electronic laboratory information system, and voice recognition dictation software. Placental specimens requiring photography, sampling for ancillary studies, or immediate pathologist's consultation were excluded. We calculated average and standard deviation of grossing times for each PA, overall average grossing time, and 95% confidence interval using a mixed linear regression model. We analyzed the impact of PA job experience, degree obtained, and number of blocks prepared on overall average in a multivariate analysis. The mean grossing times for each PA ranged from 11.0 (standard deviation [sd] = 2.0) to 17.8 (sd = 4.5) minutes. The overall average grossing time was 14.5 minutes, with a 95% confidence interval of 11.7 to 17.3 minutes. In multivariate analysis, an increase in the number of blocks prepared was significantly associated with longer overall average grossing time. If 4 blocks were prepared consistently, the model predicted a slightly lower overall average of 13.3 minutes, with a 95% confidence interval of 10.9 to 15.7 minutes. To our knowledge, our study is the first to objectively report time required for PAs to perform gross examinations of routine second and third trimester singleton placentas. The methodology of our study is replicable and can be applied to other specimen types and laboratory settings. Previously, estimated grossing times for specimens were primarily based on retrospective surveys, which were susceptible to recall errors and subjectivity. However, our study demonstrates objective data collection is achievable. Furthermore, the data collected from this study offer valuable insights into the accuracy of previous and current pathology workload models for second and third trimester singleton placentas.

ApOL-Application Oriented Workload Model for Digital Human Models for the Development of Human-Machine Systems

Since musculoskeletal disorders are one of the most common work-related diseases for assemblers and machine operators, it is crucial to find new ways to alleviate the physical load on workers. Support systems such as exoskeletons or handheld power tools are promising technology to reduce the physical load on the humans. The development of such systems requires consideration of the interactions between human and technical systems. The physical relief effect of the exoskeleton can be demonstrated in experimental studies or by simulation with the digital human model (DHM). For the digital development of these support systems, an application-oriented representation of the workload is necessary. To facilitate digital development, an application-oriented workload model (ApOL model) of an overhead working task is presented. The ApOL model determines the load (forces, torques) onto the DHM during an overhead screw-in task using a cordless screwdriver, based on experimental data. The ApOL model is verified by comparing the simulated results to the calculated values from a mathematical model, using experimental data from three participants. The comparison demonstrates successful verification, with a maximum relative mean-absolute-error (rMAE) of the relevant load components at 11.4%. The presented ApOL model can be utilized to assess the impact of cordless screwdriver design on the human workload and facilitate a strain-based design approach for support systems e.g., exoskeletons.

Moving Toward Equity and Transparency in NP Faculty Workload: It's More Than Just Credits.

Nurse practitioner (NP) programs rely on a highly educated and experienced faculty to deliver the curriculum. Full-time NP faculty are expected to be doctorally prepared, with responsibilities for instruction, an active clinical practice, scholarship, and service. Although the majority of faculty workload comes from effort in instruction, there is significant variability in how faculty workload is assigned. Absent a national model, schools of nursing are challenged to develop workload models for NP faculty to allocate effort for the full scope of work that NP faculty are expected to perform to meet requirements for academic credentialing bodies, licensure, and academic promotion. This article examines current practices in NP workload allocation and explores ways to allocate effort to the domains of work integral to NP education. NP faculty workload models should align with academic promotion expectations to encourage equity and transparency.

Workload Characterization and Classification: A Step Towards Better Resource Utilization in a Cloud Data Center

Advancements in virtualization technology have led to better utilization of existing infrastructure. It allows numerous virtual machines with different workloads to coexist on the same physical server, resulting in a pool of server resources. It is critical to understand enterprise workloads to correctly create and configure existing and future support in such pools. Managing resources in a cloud data center is one of the most difficult tasks. The dynamic nature of the cloud environment, as well as the high level of uncertainty, has created these challenges. These applications’ diverse Quality of Service (QoS) requirements make data center management difficult. Accurate forecasting of future resource demand is required to meet QoS needs and ensure better resource utilization. Consequently, data center workload modeling and categorization are needed to meet software quality solutions cost-effectively. This paper uses traces of Bitbrain’s data to characterize and categorize workload. Clustering (K Means and Gaussian mixture model) and Classification strategies (K Nearest Neighbors, Logistic Regression, Decision Trees, Random Forest, and Support Vector Machine) characterize and model the workload traces. K Means shows better results as compared to GMM when compared to the Calinski Harabasz index and Davies-Bouldin score. The results showed that the Decision Tree achieves the maximum accuracy of 99.18%, followed by K Nearest Neighbor (KNN), Random Forest (RF), Support Vector Machine (SVM) Logistic Regression (LR), Multi-Layer Perceptron (MLP), and Back Propagation Neural Networks.

Evaluation of a New Faculty Teaching Workload Model.

Creating an equitable faculty workload model is an ongoing concern. This research evaluated the effectiveness of and satisfaction with a new faculty teaching workload model 1 year after implementation. Data were collected through secondary analysis of faculty assignment spreadsheets, online survey of all full-time nursing faculty members, online survey of college of nursing administrators, and financial analysis. Individual faculty teaching loads were not consistent with the workload model. Tenure-track faculty members were assigned higher workloads than the model. Faculty members strongly preferred to have input into their schedule. Both faculty members and administrators identified strengths and opportunities for the model. Creating equitable faculty assignments is complex. Administrators and faculty members need to establish a mutual understanding of the process used to calculate equitable workloads and protect time for service and scholarship commensurate with faculty rank.

Clinical Pharmacy Faculty Provision of Direct Patient Care, Challenges, and Opportunities.

The quadripartite mission of clinical track faculty members involves research, teaching, services, and direct patient care. However, the extent of faculty involvement in direct patient care remains a challenge. Thus, the study's objective is to evaluate the effort spent on direct patient care by clinical faculty of pharmacy schools in Saudi Arabia (S.A.) and identify factors that hinder or facilitate providing direct patient care services. This multi-institutional, cross-sectional questionnaire study conducted between July 2021 and March 2022 involved several pharmacy schools' clinical pharmacy faculty members in S.A. The primary outcome was the percentage of time/effort spent on patient care services and other academic responsibilities. The secondary outcomes were the factors affecting the effort spent on direct patient care and the barriers preventing the provision of clinical services. A total of 44 faculty members took the survey. The percentage of effort spent on clinical education was highest at a median (IQR) of 37.5 (30, 50), followed by that spent on patient care [19 (10, 28.75)]. The percentages of effort spent on education and the length of academic experience were negatively associated with efforts spent on direct patient care. The most commonly reported barrier affecting fulfilling patient care duties was the lack of a clear practice policy (68%). Although most clinical pharmacy faculty members were involved in direct patient care, half of them devoted only 20% or less of their time to it. An effective effort allocation for clinical faculty duties will require the development of a clinical faculty workload model that sets realistic expectations about the time spent on clinical and non-clinical duties.

The Invisible Workload of Open Research

It is acknowledged that conducting open research requires additional time and effort compared to conducting ‘closed’ research. However, this additional work is often discussed only in abstract terms, a discourse which ignores the practicalities of how researchers are expected to find the time to engage with these practices in the context of their broader role as multifaceted academics. In the context of a sector that is blighted by stress, burnout, untenable workloads, and hyper-competitive pressures to produce, there is a clear danger that additional expectations to engage in open practices add to the workload burden and increase pressure on academics even further. In this article, the theories of academic capitalism and workload creep are used to explore how workload models currently exploit researchers by mismeasuring academic labour. The specific increase in workload resulting from open practices and associated administration is then outlined, including via the cumulative effects of administrative burden. It is argued that there is a high chance that without intervention, increased expectations to engage in open research practices may lead to unacceptable increases in demands on academics. Finally, the individual and systematic responsibilities to mitigate this are discussed.

Team workload: Review and conceptualization

Since many tasks are performed through teamwork, team workload becomes a highly concerned issue over time. However, there is still conceptual confusion. The purpose of this paper is to provide a clear and systematic model of the team workload. From the view of resource requirement, system, and interaction perspectives, models and conceptual framework of the team workload are summarized. With a broad perspective, a nominal definition of team workload is proposed as the total effort spent for processes involved in the execution of team tasks. Subsequently, the components and indicators of teamwork are identified from various team processes. This paper constructs team workload in four dimensions, task-focused activities, task-induced team activities, situational response, and team maintenance and development. Finally, the proposed model is compared with the existing models.

Cloud Workload Turning Points Prediction via Cloud Feature-Enhanced Deep Learning

Cloud workload turning point is either a local peak point standing for workload pressure or a local valley point standing for resource waste. Predicting such critical points is important to give warnings to system managers to take precautionary measures aimed at achieving high resource utilization, quality of service (QoS), and profit of the investment. Existing researches mainly focus on point value prediction only, whereas trend-based turning point prediction is not considered. This is partly due to the fact that traditional trend prediction methods have a weak ability to represent the cloud features, which means that they cannot describe the highly-variable cloud workloads time series. This paper introduces a novel cloud workload turning point prediction approach based on cloud feature-enhanced deep learning. Firstly, we establish a turning point prediction model of cloud server workload by taking cloud workload features into consideration. Then, a cloud feature-enhanced deep learning model is designed for workload turning point prediction. Experiments on the most famous Google cluster demonstrate the effectiveness of our model compared with state-of-the-art models. To the best of our knowledge, this paper is the first systematic research on turning point-based trend prediction of cloud workload time series by cloud feature-enhanced deep learning.

Benchmark Report on Utilization of Faculty Workload Models in Colleges and Schools of Pharmacy

This commentary offers a call to action to develop equity-minded, evidence-based faculty workload policies and practices within colleges and schools of pharmacy. The University of Maryland School of Pharmacy sponsored an investigation to characterize and compare peer schools’ models for measuring and using faculty workload data. An external consulting group selected 28 colleges and schools of pharmacy based on characteristics similar to the University of Maryland School of Pharmacy and collected information, feedback, and data on how these programs assessed faculty workload. Exploratory emails and phone interviews were used to collect these data. Nine of the 28 programs participated in additional follow-up discussions. These interviews identified common themes, although there was wide variability in design and implementation of workload models, even among comparable institutions. These findings align with the national Faculty Workload and Rewards Project that explored how faculty workload models can perpetuate inequities and undermine productivity, satisfaction, and retention.

Resource Management in Cloud and Cloud-influenced Technologies for Internet of Things Applications

The trend of adopting Internet of Things (IoT) in healthcare, smart cities, Industry 4.0, and so on is increasing by means of cloud computing, which provides on-demand storage and computation facilities over the Internet. To meet specific requirements of IoT applications, the cloud has also shifted its service offering platform to its next-generation models, such as fog, mist, and dew computing. As a result, the cloud and IoT have become part and parcel of smart applications that play significant roles in improving the quality of human life. In addition to the inherent advantages of advanced cloud models, to improve the performance of IoT applications further, it is essential to understand how the resources in the cloud and cloud-influenced platforms are managed to support various phases in the end-to-end IoT deployment. Considering this importance, in this article, we provide a brief description, a systematic review, and possible research directions on every aspect of resource management tasks, such as workload modeling, resource provisioning, workload scheduling, resource allocation, load balancing, energy management, and resource heterogeneity in such advanced platforms, from a cloud perspective. The primary objective of this article is to help early researchers gain insight into the underlying concepts of resource management tasks in the cloud for IoT applications.