Resource-Efficient Orchestration for Heterogeneous Serverless Computing with Harmonized Effectiveness and Practicability

A public cloud platform offers economy of scale, ease of management, and elasticity to solutions. In addition, regulatory compliance and security must be assured for solutions handling sensitive data, such as student and healthcare data. With the steep rise in data breaches at large enterprises, it is a requirement to emphasize the security, privacy, and compliance of cloud-delivered solutions that hold personally identifiable information (PII). An example of a solution in need of such assurances is an education and learning-related analytics service that handles confidential student data on a public cloud platform. In this paper, we propose an approach for managing the security and privacy of an education and learning-analytics solution on a public cloud platform while assuring compliance with the Family Educational Rights and Privacy Act (FERPA). We also propose a new agile deployment approach that is both rapid and automatic. A prototype of a learning-analytics solution was implemented on a SoftLayer public cloud, and the new deployment method was evaluated in comparison with existing methods.

Project Pythia is the educational arm of the Pangeo community, and provides a growing collection of community driven and developed training resources that help geoscientists navigate the Pangeo ecosystem, and the myriad complex technologies essential for today’s Big Data science challenges. Project Pythia began in 2020 with the support of a U.S. NSF EarthCube award. Much of the initial effort focused on Pythia Foundations: a collection of Jupyter Notebooks that covered essential topics such as Python language basics; managing projects with GitHub; authoring and using “binderized” Jupyter Notebooks; and many of Pangeo’s core packages such as Xarray, Pandas, and Matplotlib. Building upon Foundations, the Pythia community turned its attention toward creating Pythia Cookbooks: exemplar collections of recipes for transforming raw ingredients (publicly available, cloud-hosted data) into scientifically useful results. Built from Jupyter Notebooks, Cookbooks are explicitly tied to reproducible computational environments and supported by a rich infrastructure enabling collaborative authoring and automated health-checking – essential tools in the struggle against the widespread notebook obsolescence problem. Open-access, cloud-based Cookbooks are a democratizing force for growing the capacity of current and future geoscientists to practice open science within the rapidly evolving open science ecosystem. In this talk we outline our vision of a sustainable, inclusive open geoscience community enabled by Cookbooks. With further support from the NSF, the Pythia community will accelerate the development and broad buy-in of these resources, demonstrating highly scalable versions of common analysis workflows on high-value datasets across the geosciences. Infrastructure will be deployed for performant data-proximate Cookbook authoring, testing, and use, on both commercial and public cloud platforms. Content and community will expand through annual workshops, outreach, and classroom use, with recruitment targeting under-served communities. Priorities will be guided by an independent steering board; sustainability will be achieved by nurturing a vibrant, inclusive community backed by automation that lowers barriers to participation.

Recommender systems play a critical role in e-commerce, streaming media, and other fields, where user engagement is vital to platform success. These systems, however, face challenges like the cold start problem, which limits recommendation accuracy for new users and items due to sparse data. Although multi-armed bandit (MAB) algorithms are considered effective for balancing exploration and exploitation, existing studies lack simulations of algorithm performance in scenarios where recommender systems indirectly influence user choices, limiting insights for performance improvements in real-world applications. This study explores the performance of key MAB algorithmsExplore-Then-Commit (ETC), Upper Confidence Bound (UCB), and Thompson Sampling (TS)under a novel framework that models user behavior as an unreliable intermediate using Markov state transitions. This Markov perturbation-based approach provides a foundation for indirectly influencing user decisions. Two algorithmic frameworks are examined: one considering only the recommendation system's influence on user choices, and another also considering user feedback as a dynamic influence on state transition probabilities. Results show that, while considering user feedback increases adaptability, it also causes greater volatility. TS achieves the lowest cumulative regret with strong stability, while Asymptotic-UCB achieves a better balance between exploration and exploitation than standard UCB, with consistent adaptability under varying feedback. In contrast, ETC demonstrates significant volatility, highlighting its limitations in feedback-driven scenarios. These findings provide valuable insights for enhancing MAB algorithms in real-world cold start recommendation applications, especially where systems influence user choices indirectly.

Advances in the field of cloud computing and networking have led to rapid development and market growth in areas such as online retail, gaming and healthcare. In the field of advanced manufacturing however, the impact has been significantly lesser than expected due to limitations in cloud platforms for fostering community engagement. To address this problem, we study a new cloud-based architecture that provides Platform-asa-Service (PaaS) management capabilities to the manufacturing community for delivering Software-as-a-Service (SaaS) “Apps” to their customers. Our architecture aims at supporting an “App Marketplace” that thrives on agile development, organic collaboration and scalable sales of next generation manufacturing Apps requiring high-performance simulation and modeling. Towards realizing the vision of the above architecture, our paper involves investigation and implementation of an Ontology Service that interoperates with other common web services related to resource brokering and accounting. Our Ontology Service uses principles of mapping and merging to translate a manufacturing App's collaboration requirements to suitable resource specifications on public cloud platforms. Integrated resultant ontology can be queried to provision the required resource parameters such as amount of memory/storage, number of processing units, and network protocol configurations needed for deployment of an App. We validate the effectiveness of our Ontology Service using the Protege framework in a pilot testbed of a real-world “WheelSim” App in the NSF GENI Cloud platform. Our ontology integration results show benefits to an App developer in terms of: optimal user experience, lower design time and lower cost/simulation.

The ongoing need to protect key nodes of network infrastructure has been a pressing issue since the outburst of modern Internet threats. This paper presents ideas on building a novel network-based intrusion prevention system combining the advantages of different types of latest intrusion detection systems. Special attention is also given to means of traffic data acquisition as well as security policy decision and enforcement possibilities. With regard to recent trends in PaaS and SaaS, common deployment specific for private and public cloud platforms is considered.

As cloud computing thrives, many small organizations are joining a public cloud to take advantage of its multiple benefits. Cloud computing is cost efficient, i.e., cloud user can reduce spending on technology infrastructure and have easy access to their information without up-front or long-term commitment of resources. Moreover, a cloud user can dynamically grow and shrink the resources provisioned to an application on demand. Despite those benefits, cyber security concern is the main reason many large organizations with sensitive information such as the Department of Defense have been reluctant to join a public cloud. This is because different public cloud users share a common platform such as the hypervisor. A common platform intensifies the well-known problem of cyber security interdependency. In fact, an attacker can compromise a virtual machine (VM) to launch an attack on the hypervisor which if compromised can instantly yield the compromising of all the VMs running on top of that hypervisor. Therefore, a user that does not invest in cyber security imposes a negative externality on others. This research uses the mathematical framework of game theory to analyze the cause and effect of interdependency in a public cloud platform. This work shows that there are multiple possible Nash equilibria of the public cloud security game. However, the players use a specific Nash equilibrium profile depending on the probability that the hypervisor is compromised given a successful attack on a user and the total expense required to invest in security. Finally, there is no Nash equilibrium in which all the users in a public cloud will fully invest in security.

As we come to terms with various big data challenges, one vital issue remains largely untouched. That is service level agreement (SLA) management to deliver strong Quality of Service (QoS) guarantees for big data analytics applications (BDAA) sharing the same underlying infrastructure, for example, a public cloud platform. Although SLA and QoS are not new concepts as they originated much before the cloud computing and big data era, its importance is amplified and complexity is aggravated by the emergence of time-sensitive BDAAs such as social network-based stock recommendation and environmental monitoring. These applications require strong QoS guarantees and dependability from the underlying cloud computing platform to accommodate real-time responses while handling ever-increasing complexities and uncertainties. Hence, the over-reaching goal of this PhD research is to develop novel simulation, modelling and benchmarking tools and techniques that can aid researchers and practitioners in studying the impact of uncertainties (contention, failures, anomalies, etc.) on the final SLA and QoS of a cloud-hosted BDAA.

Web services have increasingly begun to rely on public cloud platforms. The virtualization technologies employed by public clouds can however trigger contention between virtual machines (VMs) for shared physical machine (PM) resources thereby leading to performance problems for the Web service. Past studies have exploited PM level performance metrics such as Clock Cycles Per Instruction to detect such platform induced performance interference. Unfortunately, public cloud customers do not have access to such metrics. They can typically only access VM-level metrics and application level metrics such as transaction response times and such metrics alone are often not useful for detecting inter-VM contention. This poses a difficult challenge to Web service operators for detecting and managing platform induced performance interference issues inside the cloud. We propose a machine learning based interference detection technique to address this problem. The technique applies collaborative filtering to predict whether a given transaction being processed by a Web service is suffering adversely from interference. The results can then be used by a management controller to trigger remedial actions, e.g., reporting problems to the system manager or switching cloud providers. Results using a realistic Web benchmark show that the approach is effective. The most effective variant of our approach is able to detect about 96% of performance interference events with almost no false alarms.

The operational cost minimization in distributed clouds via community-aware user data placements of social networks

Serverless computing has fundamentally transformed cloud technology by enabling developers to deploy applications without the burden of managing server infrastructure. This paper delves into the serverless offerings from major cloud providers—AWS (Amazon Web Services), Azure (Microsoft Azure), and GCP (Google Cloud Platform)—and explores how each contributes to this paradigm shift. The discussion highlights the core advantages of serverless computing, including cost efficiency through a pay-as-you-go model, automatic scaling that adjusts to changing demands, and reduced operational complexity, which allows developers to focus more on coding and less on infrastructure concerns. However, serverless computing is not without its challenges. Notable issues include "cold start" delays, where initial function executions may experience latency, impacting overall performance. Additionally, the shared nature of serverless environments raises potential security concerns, and the limited visibility into the execution environment can make debugging and optimization more complex. The paper also considers future trends in serverless computing across AWS, Azure, and GCP. It discusses emerging approaches such as combining serverless with traditional computing models to create hybrid architectures that leverage the strengths of both. Furthermore, it examines ongoing advancements aimed at improving support for complex, stateful applications, seeking to address current limitations and enhance the capabilities of serverless platforms. This exploration provides a detailed look at the current state of serverless computing and anticipates future developments, focusing on how AWS, Azure, and GCP are shaping the evolution of this technology

As deep learning technology paves its way, real-world applications that make use of it become popular these days. Edge computing architecture is one of the service architectures to realize the deep learning based service, which makes use of the resources near the data source or client. In Edge computing architecture it becomes important to manage resource usage, and there is research on optimization of deep learning, such as pruning or binarization, which makes deep learning models more lightweight, along with the research for the efficient distribution of workloads on cloud or edge resources. Those are to reduce the workload on edge resources. In this paper, a usage optimization method with batch and model management is proposed. The proposed method is to increase the utilization of GPU resource by modifying the batch size of the input of an inference application. To this end, the inference pipelines are identified to see how the different kinds of resources are used, and then the effect of batch inference on GPU is measured. The proposed method consists of a few modules, including a tool for batch size management which is able to change a batch size with respect to the available resources, and another one for model management which supports on-the-fly update of a model. The proposed methods are implemented on a real-time video analysis application and deployed in the Kubernetes cluster as a Docker container. The result shows that the proposed method can optimize the usage of edge resources for real-time video analysis deep learning applications.

During the past few years, serverless computing has changed the paradigm of application development and deployment in the cloud and edge due to its unique advantages, including easy administration, automatic scaling, built-in fault tolerance, etc. Nevertheless, serverless computing is also facing challenges such as long latency due to the cold start. In this paper, we present an in-depth performance analysis of cold start in the serverless framework and propose HotC, a container-based runtime management framework that leverages the lightweight containers to mitigate the cold start and improve the network performance of serverless applications. HotC maintains a live container runtime pool, analyzes the user input or configuration file, and provides available runtime for immediate reuse. To precisely predict the request and efficiently manage the hot containers, we design an adaptive live container control algorithm combining the exponential smoothing model and Markov chain method. Our evaluation results show that HotC introduces negligible overhead and can efficiently improve the performance of various applications with different network traffic patterns in both cloud servers and edge devices.

Gold is a precious metal with high resale value, often considered a safe investment as its price typically rises with inflation, attracting investors. However, even slight changes in gold prices can have significant impacts. To build an accurate forecasting model, this study applies and compares Gated Recurrent Unit (GRU), Bidirectional Gated Recurrent Unit (Bi-GRU), Long Short-Term Memory (LSTM), and Bidirectional Long Short-Term Memory (Bi-LSTM) algorithms on global gold prices. GRU and LSTM are recurrent neural networks designed to capture patterns in sequential data, where GRU uses a simplified gating mechanism to retain essential information, and LSTM, with its more complex gates, helps manage long-term dependencies in data. Bi-GRU and Bi-LSTM process data bidirectionally, capturing context from both past and future sequences for better prediction accuracy. This research uses data from Yahoo Finance (01-01-2014 to 12-06-2024) and experiments with optimization techniques (Adam, AdamW, Adamax, and Nadam), batch sizes (8, 16, and 32), time steps (10, 20, and 30), and a learning rate of 0.0001, trained for 1000 epochs with checkpoints and early stopping. Bi-GRU with Nadam, batch size 8, and 20 time steps proved most effective, with MSE of 4.1153, RMSE of 2.0286, MAE of 1.5881, and MAPE of 0.8857%. Forecasts using this model predict a 20-day decline in gold prices.

Abstract. This study presents a comparative analysis of different neural network models, including convolutional neural networks (CNN), long short-term memory (LSTM) and gated recurrent units (GRUs), with regard to predicting discharge within ungauged basins in Hesse, Germany. All models were trained on 54 catchments with 28 years of daily meteorological data, either including or excluding 11 static catchment attributes. The training process for each model scenario combination was repeated 100 times using a Latin hypercube sampler for hyperparameter optimisation with batch sizes of 256 and 2048. The evaluation was carried out using data from 35 additional catchments (6 years) to ensure predictions in basins that were not part of the training data. This evaluation assessed predictive accuracy and computational efficiency concerning varying batch sizes and input configurations and conducted a sensitivity analysis of dynamic input features. The findings indicated that all examined artificial neural networks demonstrated significant predictive capabilities, with a CNN model exhibiting slightly superior performance, closely followed by LSTM and GRU models. The integration of static features was found to improve performance across all models, highlighting the importance of feature selection. Furthermore, models utilising larger batch sizes displayed reduced performance. The analysis of computational efficiency revealed that a GRU model was 41 % faster than the CNN model and 59 % faster than the LSTM model. Despite a modest disparity in performance among the models (<3.9 %), the GRU model's advantageous computational speed rendered it an optimal compromise between predictive accuracy and computational demand.

The emergence of cloud computing brought the opportunity to use large-scale compute infrastructures for a broad spectrum of applications and users. As the cloud paradigm gets attractive for the elasticity'' in resource usage and associated costs (the users only pay for resources actually used), cloud applications still suffer from the high latencies and low performance of cloud storage services. Enabling high-throughput massive data processing on cloud data becomes a critical issue, as it impacts the overall application performance. In this paper we address the above challenge at the level of the cloud storage. We introduce a concurrency-optimized data storage system which federates the virtual disks associated to VMs. We demonstrate the performance of our solution for efficient data-intensive processing on commercial clouds by building an optimized prototype MapReduce framework for Azure that leverages the benefits of our storage solution. We perform extensive synthetic benchmarks as well as experiments with real-world applications: they demonstrate that our solution brings substantial benefits to data intensive applications compared to approaches relying on state-of-the-art cloud object storage.