Runtime Variability Research Articles

Variability Management in Self-Adaptive Systems through Deep Learning: A Dynamic Software Product Line Approach

Self-adaptive systems can autonomously adjust their behavior in response to environmental changes. Nowadays, not only can these systems be engineered individually, but they can also be conceived as members of a family based on the approach of dynamic software product lines. Through systematic mapping, we build on the identified gaps in the variability management of self-adaptive systems; we propose a framework that improves the adaptive capability of self-adaptive systems through feature model generation, variation point generation, the selection of a variation point, and runtime variability management using deep learning and the monitor–analysis–plan–execute–knowledge (MAPE-K) control loop. We compute the permutation of domain features and obtain all the possible variation points that a feature model can possess. After identifying variation points, we obtain an adaptation rule for each variation point of the corresponding product line through a two-stage training of an artificial neural network. To evaluate our proposal, we developed a test case in the context of an air quality-based activity recommender system, in which we generated 11 features and 32 possible variations. The results obtained with the proof of concept show that it is possible to manage identifying new variation points at runtime using deep learning. Future research will employ generating and building variation points using artificial intelligence techniques.

Agamotto: A Performance Optimization Framework for CNN Accelerator With Row Stationary Dataflow

We propose a software/hardware co-design framework called Agamotto for the complete design automation and performance optimization of the row stationary-based CNN accelerator. We design a scalable accelerator template whose critical design parameters can be configured. Based on the hardware template, Agamotto estimates the performance of the numerous possible hardware implementations for the target FPGA device and CNN model using the latency modeling tool. It chooses the best hardware design and generates the instructions and optimal runtime variables for each target CNN layer. As a result, Agamotto can generate the best hardware design within 61.67 seconds, achieving up to 2.8x higher hardware utilization than the original accelerator. In addition, experimental results show that the performance estimation is accurate, showing only 4.8% difference against the FPGA runtime for the end-to-end CNN model execution. The accelerator implemented on the Xilinx VCU118 evaluation board achieves 402 giga operations per second (GOPS) at 200 MHz, resulting in 13 frames per second (FPS) for the end-to-end execution of VGG-16. It is flexible enough to run more complex CNN models such as ResNet-50 and DarkNet-53, achieving 29.3 FPS and 16.9 FPS, respectively.

Optimizing throughput of Seq2Seq model training on the IPU platform for AI-accelerated CFD simulations

Intelligence Processing Units (IPU) have proven useful for many AI applications. In this paper, we evaluate them within the emerging field of AI for simulation, where traditional numerical simulations are supported by artificial intelligence approaches. We focus specifically on a program for training machine learning models supporting a computational fluid dynamics application. We use custom TensorFlow provided by the Poplar Software Development Kit to adapt the program for the IPU-POD16 platform and investigate its ease of use and performance scalability. Training a model on data from OpenFOAM simulations allows us to get accurate simulation state predictions in test time. We describe how to optimize multi-threading runtime options and utilize the popdist library to overcome a performance bottleneck in feeding training data to the IPU on the host side. Due to communication overheads, using data parallelism to utilize two IPUs instead of one does not improve the throughput. However, once the intra-IPU costs have been paid, the hardware capabilities for inter-IPU communication allow for good scalability. Increasing the number of IPUs from two to 16 improves the throughput from 560.8 to 2805.8 samples/s. Additionally, the experimental results show that reducing the precision of input data storage from FP32 to FP16 allows to improve training throughput by 12%, while tuning selected runtime variables, by up to 6.3%.

DaeMon: Architectural Support for Efficient Data Movement in Fully Disaggregated Systems

Resource disaggregation offers a cost effective solution to resource scaling, utilization, and failure-handling in data centers by physically separating hardware devices in a server. Servers are architected as pools of processor, memory, and storage devices, organized as independent failure-isolated components interconnected by a high-bandwidth network. A critical challenge, however, is the high performance penalty of accessing data from a remote memory module over the network. Addressing this challenge is difficult as disaggregated systems have high runtime variability in network latencies/bandwidth, and page migration can significantly delay critical path cache line accesses in other pages. This paper conducts a characterization analysis on different data movement strategies in fully disaggregated systems, evaluates their performance overheads in a variety of workloads, and introduces DaeMon, the first software-transparent mechanism to significantly alleviate data movement overheads in fully disaggregated systems. First, to enable scalability to multiple hardware components in the system, we enhance each compute and memory unit with specialized engines that transparently handle data migrations. Second, to achieve high performance and provide robustness across various network, architecture and application characteristics, we implement a synergistic approach of bandwidth partitioning, link compression, decoupled data movement of multiple granularities, and adaptive granularity selection in data movements. We evaluate DaeMon in a wide variety of workloads at different network and architecture configurations using a state-of-the-art simulator. DaeMon improves system performance and data access costs by 2.39× and 3.06×, respectively, over the widely-adopted approach of moving data at page granularity.

An Evaluation of Comprehensive Transit Improvements—TriMet’s Streamline Program

Transit performance is influenced by a variety of factors in an urban environment. Making transit more convenient and competitive with automobile travel is a key objective for the Tri-County Metropolitan Transportation District of Oregon (TriMet). TriMet’s goal is to have a “Total Transit System” that makes transit an attractive choice for riders. Portland’s Streamline program has been a significant effort toward meeting these goals. The program has resulted in operating and capital cost savings for TriMet by delaying the need to add more buses to the fleet as well as operating savings due to reductions in running time variability. Further, the way the program was implemented resulted in a greater increase in ridership than would have been achieved had the service increases been spread more evenly around the system, confirming that the BRT approach serves transit agencies effectively by concentrating improvements on corridors.

Sensitivity analysis of an ammonium salt formation model applied to pollutant removal in marine diesel exhaust gases

Elevated exposure to airborne pollutants such as NOx and SOx is known to be damaging to human health. A current approach to deal with such harmful gases is to trap them in ammonium salt particles. The present study presents the sensitivity analysis of the aerosol model for ammonium salt particle formation from NOx and SOx for low-temperature gas cleaning applications developed by Olenius et al. (2021). Starting from the acid gases derived from NOx and SOx (i.e. HNO3 and H2SO4), the model simulates the particle growth phenomena as the acids react with ammonia (NH3). This work presents, for the first time, a global sensitivity analysis of the aerosol model uncertainty. The first- and total-order effects of five different input variables on model outputs such as particle size distribution, pollutant removal effectiveness, ammonia slip, and total run time are reported. Furthermore, the range of input parameters for which the model is tested is made to emulate the conditions experienced by two-stroke marine diesel engine ships. Sources of uncertainty are reviewed in detail to provide a complete view of the knowledge gaps in the particle conversion process. For the conditions studied, we report that variations in particle sizes are influenced by HNO3, H2SO4 and temperature. Similarly, the degree of ammonia slip was observed to be driven by temperature and the ammonia ratio. Additionally, the removal efficiency of HNO3 was reported to be very high (above 99%) for the vast majority of conditions tested, and was not significantly influenced by the concentration of H2SO4. Finally, the model run time variability was observed to depend mainly on variations in temperature, relative humidity and the ammonia ratio.

Reinforcement learning and A* search for the unit commitment problem

Previous research has combined model-free reinforcement learning with model-based tree search methods to solve the unit commitment problem with stochastic demand and renewables generation. This approach was limited to shallow search depths and suffered from significant variability in run time across problem instances with varying complexity. To mitigate these issues, we extend this methodology to more advanced search algorithms based on A* search. First, we develop a problem-specific heuristic based on priority list unit commitment methods and apply this in Guided A* search, reducing run time by up to 94% with negligible impact on operating costs. In addition, we address the run time variability issue by employing a novel anytime algorithm, Guided IDA*, replacing the fixed search depth parameter with a time budget constraint. We show that Guided IDA* mitigates the run time variability of previous guided tree search algorithms and enables further operating cost reductions of up to 1%.

RuVa: A Runtime Software Variability Algorithm

Context-aware and smart systems that require runtime reconfiguration to cope with changes in the environment increasingly demand variability management mechanisms that can address runtime concerns. In recent years, we have witnessed new dynamic variability solutions using dynamic software product line (DSPL) approaches. However, while few solutions proposed so far have addressed the need to add, change and remove variants dynamically, none of them provide a way to check the constraints between features at runtime. Because all SAT solvers perform variability constraint checking in off-line mode, we suggest in this ongoing research paper the integration of RuVa, a runtime variability algorithm, with the FaMa tool suite to check feature constraints dynamically before a new feature is added or an existing feature is removed. This research suggests a novel approach to modifying the variability model of context-aware systems dynamically and check the feature constraints on the fly.We integrate our solution with a SAT solver that can be invoked at runtime by a cyber-physical system.We validate the effectiveness and performance of the proposed algorithm using simulations.We also provide a proof-of-concept for updating the configuration of a robot’s variability model based on contextual changes.

Component-based Runtime Variability Model for Safe Dynamic Reconfiguration

Component-based Runtime Variability Model for Safe Dynamic Reconfiguration

Quantifying the effects of running time variability on the capacity of rail corridors

Traffic variability is well known to have a substantial effect on railway capacity. Varying train running and stopping times entail larger train separations and yield non-usable time slots in train timetables and operations. In this paper, we aim to assess the effects of uncertainty in running and dwell times on the capacity of railway corridors in long term planning of rail traffic. Our main focus are commuter and metro systems, where the effects of fluctuations of running and stopping times are particularly pronounced due to dense operations.To analyze the effects of variability in train operations, we propose a new stochastic approach based on a serial queuing network with finite-capacity service stations. The corridor is modeled as a sequence of heavily correlated service stations representing line segments and stations, for which effective throughput, distributions of train running times and service quality are calculated. The performance of the model is tested in a case study for the central link of the mass transit system in Cologne. In addition, an outlook on how the model can be extended to general heavy-rail corridors with different types of train services is provided.

Straggler Mitigation at Scale

Runtime performance variability has been a major issue, hindering predictable and scalable performance in modern distributed systems. Executing requests or jobs redundantly over multiple servers have been shown to be effective for mitigating variability, both in theory and practice. Systems that employ redundancy has drawn significant attention, and numerous papers have analyzed the pain and gain of redundancy under various service models and assumptions on the runtime variability. This paper presents a cost (pain) vs. latency (gain) analysis of executing jobs of many tasks by employing replicated or erasure coded redundancy. The tail heaviness of service time variability is decisive on the pain and gain of redundancy and we quantify its effect by deriving expressions for cost and latency. Specifically, we try to answer four questions: 1) How do replicated and coded redundancy compare in the cost vs. latency tradeoff? 2) Can we introduce redundancy after waiting some time and expect it to reduce the cost? 3) Can relaunching the tasks that appear to be straggling after some time help to reduce cost and/or latency? 4) Is it effective to use redundancy and relaunching together? We validate the answers we found for each of these questions via simulations that use empirical distributions extracted from a Google cluster data.

A Device-to-System Perspective Regarding Self-Heating Enhanced Hot Carrier Degradation in Modern Field-Effect Transistors: A Topical Review

As foreseen by Keyes in the late 1960s, the self-heating effect has emerged as an important concern for device performance, output power density, run-time variability, and reliability of modern field-effect transistors. The self-heating effect is aggravated as the device footprint scales down for higher level of integration (low-power devices) or as the devices are operated in ultrahigh voltage regimes (high-power devices). In this article, we focus on the implications of self-heating on hot carrier degradation (HCD) of modern transistors by integrating within a coherent theoretical framework a broad range of experimental data scattered in the literature. We explain why system integration exacerbates transistor self-heating, while high-frequency digital operation ameliorates it, suggesting an opportunity for co-optimization. We conclude this article by discussing the various material–device–system design strategies to reduce HCD and suggesting open problems for further research.

Selective Flip-Flop Optimization for Reliable Digital Circuit Design

Runtime variability sources, such as bias temperature instability (BTI) and supply voltage fluctuation affect both timing and functionality of the flip-flops inside a VLSI circuit. In this paper, we propose a method to improve the timing and reliability of the VLSI circuits by optimizing the flip-flops for resiliency against aging and supply voltage fluctuation. In the proposed selective reliability optimization method, we first extend the standard cell libraries by adding optimized versions of the flip-flops designed for better resiliency against severe BTI impact and/or supply voltage fluctuation. Then, we optimize the VLSI circuit by replacing the aging-critical and voltage-drop-critical flip-flops (VC) of the circuit with the reliability-optimized versions to improve the timing and the reliability of the entire circuit in a cost-effective way. The simulation results show that incorporating the optimized flip-flops in a processor can prolong the lifetime of the processor by 36.9% compared to the original design, which translates into better reliability. This is achieved with negligible leakage overhead (less than 0.1% on the processor) and no area overhead which facilitates the integration of the proposed method in the standard VLSI design flow.

Toward Supporting Unplanned Dynamic Changes of Service-Based Business Processes

Service-oriented architecture (SOA) has become an application development paradigm widely recognized both in academia and in industry. Although SOA-based applications may vertically implement business processes through the composition of loosely coupled services, they have to face frequent changes, such as unavailability of service (due to the uncontrollable, dynamic, and distributed environments) or dynamic replacement of service (due to specific user requirements). This indicates that the service compositions are expected to be adaptable enough to cater to such changing situations. In this paper, we propose an approach to supporting unplanned dynamic changes of service compositions by combining variability management and dynamic binding. The proposed approach introduces the concept of abstract proxy services in a variability-supporting service composition language, namely VxBPEL, and provides a mechanism to support variation design and dynamic binding for unplanned changes at run time. To realize the proposed approach, we have developed a service composition engine that supports abstract proxy services and their run-time replacement via service discovery or user intervention. Finally, a case study has been conducted to demonstrate the feasibility of the proposed approach and quantify the performance overhead resulting from runtime variability management and dynamic binding. The experimental results show that the proposed approach overcomes the limitation of imperative variability-based approaches in handling unplanned dynamic changes and further enhances the dynamic adaptability of VxBPEL-based service compositions.

Loughborough Soccer Passing Test é reprodutível em superfície específica à prática de futebol de campo

Objetivou-se verificar a reprodutibilidade teste-reteste do Loughborough Soccer Passing Test (LSPT) em quadra coberta e campo gramado. Ademais, buscou-se comparar e correlacionar os resultados de jovens futebolistas. Para isso, dez participantes com média de idade de 14,7±0,5 anos, 173,4±9,1 cm de estatura, e 61,5±9,0 kg realizaram duas medidas de avaliação, separadas por sete dias, do protocolo do LSPT em campo gramado e quadra coberta. Como resultados, os valores de Coeficiente Correlação Intraclasse, Coeficiente de Variação, Erro Padrão de Medida e Nível de Concordância sugerem que o LSPT seja protocolo reprodutível para o campo gramado. Na quadra coberta, o Tempo Limpo (TLim) foi de 47,8±4,4s, o Tempo de Penalidades de 21,4±9,9s e o Tempo de Desempenho (TD) de 65,7±13,3s. Já no campo gramado foi de 49,3±4,2s, 29±7,4s, e 76,6±12s para Limpo, Penalidades e Desempenho, respectivamente. Não foi encontrada diferença estatística significativa para Tempo Limpo e Tempo de Desempenho (TLim: t = 0,10; p = 0,36; TD: t = 0,31; p = 0,17). No entanto, para o Tempo de Penalidades (TPenal) houve diferença entre os valores obtidos (TPenal: t = 0,14; p = 0,02). Não foram identificadas correlações estatisticamente significantes para as variáveis de TLim (r = 0,59; p = 0,07) e TPenal (r = 0,49; p = 0,15), mas, para o TD houve correlação (r = 0,64; p = 0,04). Conclui-se que o LSPT é ferramenta de avaliação de habilidade específica de passe em superfície específica à prática de futebol. Ainda, não houve diferença estatisticamente significativa no desempenho do LSPT entre superfícies distintas, embora mais penalidades foram verificadas no campo gramado. A correlação de TD no LSPT entre as diferentes superfícies sugere que os protocolos apontam para o mesmo objetivo de avaliação.

Prediction of arrival times of freight traffic on US railroads using support vector regression

Variability of travel times on the United States freight rail network is high due to large network demands relative to infrastructure capacity, especially when traffic is heterogeneous. Variable runtimes pose significant operational challenges if the nature of runtime variability is not predictable. To address this issue, this article proposes a data-driven approach to predict estimated times of arrival (ETAs) of individual freight trains, based on the properties of the train, the properties of the network, and the properties of potentially conflicting traffic on the network. The ETA prediction problem from an origin to a destination is posed as a machine learning regression problem and solved using support vector regression trained and cross validated on over two years of detailed historical data for a 140 mile section of track located primarily in Tennessee, USA. The article presents the data used in this problem and details on feature engineering and construction for predictions made across the full route. It also highlights findings on the dominant sources of runtime variability and the most predictive factors for ETA. Improvement results for ETA exceed 21% over a baseline prediction method at some locations and average 14% across the study area.

Verification of network end-to-end latencies for adaptive ethernet-based cyber-physical systems

As Cyber-Physical Systems (CPS) are evolving towards flexible and smart systems, their dependable communication becomes a decisive factor. In order to still guarantee a predictive and real-time behavior, verifying the network performance of such adaptive systems is vital. Therefore, the performance-verification has to consider the runtime variability while scaling for larger number of applications and networks in CPS. We introduce a novel performance-verification approach with integrated variability enabling the analysis of adaptive Ethernet-based CPS. It incorporates a formal model capturing all relevant characteristics for deriving safe communication bounds. Its soundness has been evaluated in an extensive automotive case study and several changing test setups targeting scalability. The results show that this integrated variability approach is superior to a common static analysis and previously utilized heuristic. In direct comparison it outperforms static analysis by up to 95 percent within the evaluated automotive system. Moreover, the results show that it scales well and provides a profound basis for analyzing larger adaptive networked systems.

Towards Dynamic Evolution of Runtime Variability Based on Computational Reflection

Given the frequently changing nature of the user requirements and environments in software systems, runtime variability in today’s software systems should be capable of evolving during execution. Computational reflection is required to facilitate accessing and customizing runtime variability during this evolution process. However, realizing this computational reflection includes various practical complexities since the runtime variability is typically neither explicitly represented in software systems nor changeable during runtime. To address this problem, this paper proposes a software architecture to support computational reflection of runtime variability, along with a corresponding causal-connection mechanism to realize the introspection and intercession (i.e. representing runtime variability model, and adding, removing, replacing variability elements and their relations). The proposed software architecture consists of a meta level that represents runtime variability model using objectification, and a base level that organizes and manipulates the implementation of variability elements via reconfiguration. The causal-connection mechanism integrated in our proposed model is designed to synchronize the representation and the implementation. Further, we developed a Reflective Runtime Variability Framework (R2VF) to support the development and operation of the systems with the reflection of runtime variability. The effectiveness and applicability of our approach has been evaluated by applying R2VF to Personal Data Resource Network.

Analysing improvements to on-street public transport systems: a mesoscopic model approach

Light rail transit and bus rapid transit have shown to be efficient and cost-effective in improving public transport systems in cities around the world. As these systems comprise various elements, which can be tailored to any given setting, e.g. pre-board fare-collection, holding strategies and other advanced public transport systems (APTS), the attractiveness of such systems depends heavily on their implementation. In the early planning stage it is advantageous to deploy simple and transparent models to evaluate possible ways of implementation. For this purpose, the present study develops a mesoscopic model which makes it possible to evaluate public transport operations in details, including dwell times, intelligent traffic signal timings and holding strategies while modelling impacts from other traffic using statistical distributional data thereby ensuring simplicity in use and fast computational times. This makes it appropriate for analysing the impacts of improvements to public transport operations, individually or in combination, in early planning stages. The paper presents a joint measure of reliability for such evaluations based on passengers’ perceived travel time by considering headway time regularity and running time variability, i.e. taking into account waiting time and in-vehicle time. The approach was applied on a case study by assessing the effects of implementing segregated infrastructure and APTS elements, individually and in combination. The results showed that the reliability of on-street public transport operations mainly depends on APTS elements, and especially holding strategies, whereas pure infrastructure improvements induced travel time reductions. The results further suggested that synergy effects can be obtained by planning on-street public transport coherently in terms of reduced travel times and increased reliability.

Effective Analysis of C Programs by Rewriting Variability

Context. Variability-intensive programs (program families) appear in many application areas and for many reasons today. Different family members, called variants, are derived by switching statically configurable options (features) on and off, while reuse of the common code is maximized. Inquiry. Verification of program families is challenging since the number of variants is exponential in the number of features. Existing single-program analysis and verification tools cannot be applied directly to program families, and designing and implementing the corresponding variability-aware versions is tedious and laborious. Approach. In this work, we propose a range of variability-related transformations for translating program families into single programs by replacing compile-time variability with run-time variability (non-determinism). The obtained transformed programs can be subsequently analyzed using the conventional off- the-shelf single-program analysis tools such as type checkers, symbolic executors, model checkers, and static analyzers. Knowledge. Our variability-related transformations are outcome-preserving, which means that the relation between the outcomes in the transformed single program and the union of outcomes of all variants derived from the original program family is equality. Grounding. We show our transformation rules and their correctness with respect to a minimal core imperative language IMP. Then, we discuss our experience of implementing and using the transformations for efficient and effective analysis and verification of real-world C program families. Importance. We report some interesting variability-related bugs that we discovered using various state-of-the-art single-program C verification tools, such as Frama-C, Clang, LLBMC.