Performance Degradation Of Applications Research Articles

Numerical Investigation of Flow Characteristics in Internally Flowing Pipes with Varying Roughness and Helical Angles of Vortex Generators

In this study, the turbulent behavior inside a pipe containing an aluminum helical vortex generator plate was investigated using computational fluid dynamics (CFD) methods, considering different plate roughness and helical angle values. Accordingly, numerical analyses of velocity and pressure distributions for a liquid flow with a constant inlet velocity of 0.5 m/s and four different helical angles were performed and compared in a computer software. To observe the effects of the roughness values of the metal plate on flow behavior, analyses were conducted for three different roughness values for each helical angle and the results were compared and interpreted. As a result of the study, the combination of a roughness value of 0.04 and a helical angle of 0° led to the highest pressure and velocity values, while the same roughness value combined with a 270° helical angle resulted in the lowest pressure and velocity values. Detailed analysis showed that helical angles (90° and 180°) presented moderate pressure and velocity values, indicating a non-linear relationship between helical angle and flow characteristics. The results demonstrate that optimizing the helical angle and roughness is crucial for enhancing the efficiency of vortex generators. This way, a design will be developed to prevent performance degradation in industrial applications that require flow efficiency and pressure management.

Radiation Damage on Silicon Photomultipliers from Ionizing and Non-Ionizing Radiation of Low-Earth Orbit Operations.

Silicon Photomultipliers (SiPMs) are single photon detectors that gained increasing interest in many applications as an alternative to photomultiplier tubes. In the field of space experiments, where volume, weight and power consumption are a major constraint, their advantages like compactness, ruggedness, and their potential to achieve high quantum efficiency from UV to NIR makes them ideal candidates for spaceborne, low photon flux detectors. During space missions however, SiPMs are usually exposed to high levels of radiation, both ionizing and non-ionizing, which can deteriorate the performance of these detectors over time. The goal of this work is to compare process and layout variation of SiPMs in terms of their radiation damage effects to identify the features that helps reduce the deterioration of the performance and develop the next generation of more radiation-tolerant detectors. To do this, we used protons and X-rays to irradiate several Near Ultraviolet High-Density (NUV-HD) SiPMs with small areas (single microcell, 0.2 × 0.2 mm2 and 1 × 1 mm2) produced at Fondazione Bruno Kessler (FBK), Italy. We performed online current-voltage measurements right after each irradiation step, and a complete functional characterization before and after irradiation. We observed that the main contribution to performance degradation in space applications comes from proton damage in the form of an increase in primary dark count rate (DCR) proportional to the proton fluence and a reduction in activation energy. In this context, small active area devices show a lower DCR before and after irradiation, and we propose light or charge-focusing mechanisms as future developments for high-sensitivity radiation-tolerant detectors.

Investigating the impact of microservice-oriented platform configurations on application performance

Effective management of containerized applications is crucial to ensuring their performance and reliability. The aim of this work was to indicate which configuration settings of the Kubernetes orchestrator have the greatest impact on microservice application performance under conditions of increased load. For each of the established configuration variants, the throughput and response time of the test application based on the microservices paradigm were measured. Research findings indicate that excessive horizontal scaling degrades application performance and that memory usage settings may play a greater role in optimizing system performance than CPU usage.

OptimML: Joint Control of Inference Latency and Server Power Consumption for ML Performance Optimization

Power capping is an important technique for high-density servers to safely oversubscribe the power infrastructure in a data center. However, power capping is commonly accomplished by dynamically lowering the server processors’ frequency levels, which can result in degraded application performance. For servers that run important machine learning (ML) applications with Service-Level Objective (SLO) requirements, inference performance such as recognition accuracy must be optimized within a certain latency constraint, which demands high server performance. In order to achieve the best inference accuracy under the desired latency and server power constraints, this paper proposes OptimML, a multi-input-multi-output (MIMO) control framework that jointly controls both inference latency and server power consumption, by flexibly adjusting the machine learning model size (and so its required computing resources) when server frequency needs to be lowered for power capping. Our results on a hardware testbed with widely adopted ML framework (including PyTorch, TensorFlow, and MXNet) show that OptimML achieves higher inference accuracy compared with several well-designed baselines, while respecting both latency and power constraints. Furthermore, an adaptive control scheme with online model switching and estimation is designed to achieve analytic assurance of control accuracy and system stability, even in the face of significant workload/hardware variations.

CARE: Cloudified Android With Optimized Rendering Platform

Due to the excellent rendering capabilities, GPUs are mainstream accelerators in the Cloud-rendering industry. However, current Cloud-rendering systems suffer from a CPU-GPU workload imbalance that not only degrades application performance but also causes a significant waste of GPU resources. Recent proposals (such as API-forwarding and c-GPU) for improving CPU-GPU balance are promising but fail to solve system-resource redundancy issues (i.e., each instance tends to occupy all resources, exceeding its requirements). Such behavior will increase CPU load and lower effective GPU utilization. To demonstrate the severity of the issue, we evaluated real-world applications and results show that in most cases, nearly 50% of resources are useless. To solve this problem, we present CARE, the first framework intended to reduce the system-level redundancy by cloudifying the system from monolithic to Cloud-native. To allow users to configure required services, CARE puts forward a functional unit called <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Configurable Android (CA)</i> . To allow multiple instances to share certain types of resources, CARE innovates <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Sharing Resource (SR)</i> . To reduce the unused services, CARE introduces <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Pruning Resources (PR)</i> . To further alleviate the CPU pressure and achieve CPU-GPU balance, we propose rShare, a system aiming at enhancing CPU effective utilization and increasing Android instance density of the Cloud-rendering platform. Based on Kubernetes, rShare divides all the CPUs into non-overlapping shared CPU pools, allocates instances to pools within milliseconds, and dynamically migrates them by tracking their QoS status. So far, CARE primarily focuses on Android systems and can handle 60 heavyweight instances (e.g., KOG (King of Glory)) on Intel®SG1. rShare can apply instance allocation within milliseconds and increase the platform density by 39.4%.

Energy saving of cluster computing by CPU frequency Tuning using genetic algorithm

Dynamic voltage and frequency scaling (DVFS) is a tool used primarily to decrease computer processor energy consumption by lowering its operational frequency. Their only downside is that they distract from the efficiency of parallel applications while operating on parallel platforms. In a heterogeneous cluster architecture, however, a genetic algorithm is being implemented and applied to model the best trade-off between energy-saving and parallel application performance degradation. The proposed algorithm selects the best frequency vector in order to accomplish these objectives by providing the same compromise. So, the objective function of the genetic algorithm at the same time gives limited energy consumption and minimum decreases in performance. The SimGrid simulator will be used for all experiments. The suggested algorithm saves the average energy by (20 %) and the application performance degrades to the limit (0.15 %).

Adaptive Feature Selection for Predicting Application Performance Degradation in Edge Cloud Environments

Adaptive Feature Selection for Predicting Application Performance Degradation in Edge Cloud Environments

Leveraging AWS SNS/SQS for Real-Time Cache Synchronization in Distributed Systems

In distributed systems, maintaining consistency across multiple caches presents a significant challenge, often leading to data inconsistencies and degraded application performance. This paper proposes and evaluates a novel approach to cache synchronization using Amazon Web Services (AWS) Simple Notification Service (SNS) and Simple Queue Service (SQS). The proposed system leverages the pub-sub model of SNS combined with the reliable message queuing of SQS to create a scalable and robust cache synchronization mechanism. We present the architecture, implementation details, and performance analysis of this system, demonstrating significant reductions in database load and improvements in response times while ensuring eventual consistency across distributed caches. Comparative analysis with alternative solutions highlights the advantages of the SNS/SQS approach in terms of scalability, reliability, and ease of implementation. This research contributes to the field of distributed systems by offering a practical, cloud-native solution to the cache synchronization problem, applicable to a wide range of applications requiring consistent data across multiple instances or services.

Optimization-Free Test-Time Adaptation for Cross-Person Activity Recognition

Human Activity Recognition (HAR) models often suffer from performance degradation in real-world applications due to distribution shifts in activity patterns across individuals. Test-Time Adaptation (TTA) is an emerging learning paradigm that aims to utilize the test stream to adjust predictions in real-time inference, which has not been explored in HAR before. However, the high computational cost of optimization-based TTA algorithms makes it intractable to run on resource-constrained edge devices. In this paper, we propose an Optimization-Free Test-Time Adaptation (OFTTA) framework for sensor-based HAR. OFTTA adjusts the feature extractor and linear classifier simultaneously in an optimization-free manner. For the feature extractor, we propose Exponential Decay Test-time Normalization (EDTN) to replace the conventional batch normalization (CBN) layers. EDTN combines CBN and Test-time batch Normalization (TBN) to extract reliable features against domain shifts with TBN's influence decreasing exponentially in deeper layers. For the classifier, we adjust the prediction by computing the distance between the feature and the prototype, which is calculated by a maintained support set. In addition, the update of the support set is based on the pseudo label, which can benefit from reliable features extracted by EDTN. Extensive experiments on three public cross-person HAR datasets and two different TTA settings demonstrate that OFTTA outperforms the state-of-the-art TTA approaches in both classification performance and computational efficiency. Finally, we verify the superiority of our proposed OFTTA on edge devices, indicating possible deployment in real applications. Our code is available at https://github.com/Claydon-Wang/OFTTA.

Microservices performance forecast using dynamic Multiple Predictor Systems

Time series forecasting has been applied to predict performance degradation in Microservice-Based Applications (MBAs). The prediction enables MBA adaptation to avoid performance degradation and maintain the customer experience. The approaches in the literature commonly perform the forecast using a single monolithic model. However, since no model is better than others for all possible scenarios, using only one increases the risk of inaccurate estimates. Thus, an alternative for improving the accuracy and robustness of the performance degradation of MBAs is to adopt an ensemble, i.e., a Multiple Predictor Systems (MPS). This paper proposes an MPS methodology that selects and combines the most suitable models to forecast test patterns from a pool of models. Experiments were carried out with 32 time series containing performance metrics commonly used for MBA adaptation. Different scenarios concerning the pool (homogeneous or heterogeneous) and the selection phase (static or dynamic) are considered. Likewise, six widely used models in the literature were employed for pool generation: ARIMA, Multilayer Perceptron, Support Vector Regression, Random Forest, Long Short-Term Memory, and eXtreme Gradient Boosting. The evaluation shows that the proposed solution maintains or improves the forecast accuracy on 26 out of 32 datasets (81.25%) compared to monolithic models. In particular, the homogeneous pool and the dynamic selection of predictors obtained very satisfactory results. Therefore, MPS provide more accurate decision-making for MBA interventions, avoiding incorrect adjustments that could degrade the customer experience. Source code, figures, and data sets are publicly available at https://github.com/gfads/mps-methodology.

Unleashing multi-scale mechanical enhancement in NiTi shape memory alloy via annular intra-laser deposition with homogenized Ti2Ni nanoprecipitates

Additive manufacturing (AM) of composition-sensitive NiTi shape memory alloys (SMAs) is hindered by uncontrollable non-equilibrium microstructures, leading to functional performance degradation and limitations in high-throughput net-shaping applications. To address these challenges, we propose a novel approach combining annular intra-laser directed energy deposition (AIL-DED) with a tailored gradient post-treatment process. This enables the fabrication of nanocomposite microstructures in NiTi SMAs with precise control over the homogenization of Ti2Ni nanoprecipitates. In-situ monitoring and characterization of thermal history, microstrain, and constitutional phases, were utilized to uncover the relationship between multi-scale structural features and mechanical responses. The interplay between laser reflectivity, energy input, and lattice thermal vibrations affects forming quality in AIL-DED. Dynamic temperature variations were studied before reaching thermal equilibrium, revealing insights into thermal and mechanical interactions during the process. These findings contribute to understanding challenges in fabricating high-quality NiTi SMAs. Examining phase transformation behavior and its correlation with processing parameters deepens understanding of microstructural evolution and provides insights for tailoring material properties. The stress relaxation and dislocation evacuation at elevated temperatures caused the expansion of lattice and interplanar spacing in the age-treated SMA's B2 matrix phase, exhibiting a stable two-step phase transformation behavior (B19′↔R↔B2). AIL-DED combined with two-stage heat treatment strategy demonstrated enhanced homogeneity of non-equilibrium microstructures, resulting in the formation of high-density, well-dispersed Ti2Ni nanoprecipitates, with a semi-coherent interface to B2 matrix. The combination of load-bearing strengthening through well-dispersed Ti2Ni nanoprecipitates and Orowan strengthening at the semi-coherent interface yielded simultaneous enhancements in nanohardness, ultimate compressive strength, compressive fracture strain, and deformation recovery, following the AIL-DED combined with two-stage heat treatment process. The fracture mechanism transitioned from a mixture of brittle-ductile behavior to predominantly ductile characteristics. This study provides valuable insights that advance additive manufacturing and enhance the functional performance of SMAs in critical applications.

QZRAM: A Transparent Kernel Memory Compression System Design for Memory-Intensive Applications with QAT Accelerator Integration

In recent decades, memory-intensive applications have experienced a boom, e.g., machine learning, natural language processing (NLP), and big data analytics. Such applications often experience out-of-memory (OOM) errors, which cause unexpected processes to exit without warning, resulting in negative effects on a system’s performance and stability. To mitigate OOM errors, many operating systems implement memory compression (e.g., Linux’s ZRAM) to provide flexible and larger memory space. However, these schemes incur two problems: (1) high-compression algorithms consume significant CPU resources, which inevitably degrades application performance; and (2) compromised compression algorithms with low latency and low compression ratios result in insignificant increases in memory space. In this paper, we propose QZRAM, which achieves a high-compression-ratio algorithm without high computing consumption through the integration of QAT (an ASIC accelerator) into ZRAM. To enhance hardware and software collaboration, a page-based parallel write module is introduced to serve as a more efficient request processing flow. More importantly, a QAT offloading module is introduced to asynchronously offload compression to the QAT accelerator, reducing CPU computing resource consumption and addressing two challenges: long CPU idle time and low usage of the QAT unit. The comprehensive evaluation validates that QZRAM can reduce CPU resources by up to 49.2% for the FIO micro-benchmark, increase memory space (1.66×) compared to ZRAM, and alleviate the memory overflow phenomenon of the Redis benchmark.

MDFU-Net: Multiscale dilated features up-sampling network for accurate segmentation of tumor from heterogeneous brain data

The existing methods for accurate brain tumor (BT) segmentation based on homogeneous datasets show significant performance degradation in actual clinical applications and lacked heterogeneous data analysis. To address these issues, we designed a deep learning-based multiscale dilated features up-sampling network (MDFU-Net) for accurate BT segmentation from heterogeneous brain data. Our method primarily uses the strength of multiscale dilated features (MDF) inside the encoder module to improve the segmentation performance. For the final segmentation, a simple yet effective decoder module is designed to process the dense spatial MDF. For experiments, our MDFU-Net is trained on one dataset and tested with another dataset in a heterogeneous environment, showing quantitative results of the Dice similarity coefficient (DC) of 62.66%, intersection over union (IoU) of 56.96%, specificity (Spe) of 99.29%, and sensitivity (Sen) of 51.98%, which were higher than those of the state-of-the-art methods. There are several reasons for the lower values of the evaluation metrics of the heterogeneous dataset, including the change in characteristics of different MRI modalities, the presence of minor lesions, and a highly imbalanced dataset. Moreover, the experimental results for a homogeneous dataset showed that our MDFU-Net achieved a DC of 82.96%, IoU of 74.94%, Spe of 99.89%, and Sen of 68.05%, which were also higher than those of the state-of-the-art methods. Our system, which is based on heterogeneous brain data as well as homogeneous brain data, can be advantageous to radiologists and medical experts.

Architecture for IoT applications based on reactive microservices: A performance evaluation

The Internet has evolved from a network interconnecting computers to a complex ecosystem integrating devices of the most varied types, and enabling the amalgamation of the physical and virtual worlds. Integrating these heterogeneous devices fosters novel services and applications that generate value-added information and actionable knowledge for the end-user. Several challenges are involved in the design and building of IoT ecosystems, which have fostered research in the field, in search of patterns, guidelines, methods and tools that support such activities. In terms of architectural patterns, the microservice architectural style has been increasingly adopted in the development of IoT applications and services. Its adoption promotes some essential properties in IoT, such as scalability and extensibility. As there are always tradeoffs involved in every architectural decision, it is important to analyze whether the benefits brought by the use of microservices do not come at the expense of some loss or degradation of application performance. Recent research has analyzed the performance interference of microservices based on edge computing applications. However, a comprehensive assessment of the performance impact of characteristics inherent to the use of reactive microservices on IoT applications is still missing in the literature. In this paper, we present an experimental evaluation of the performance of IoT applications that make use of an architecture based on reactive microservices. The architecture was proposed by our group in a previous work and was tailored for reliable IoT applications running at the edge of the network. The experiments presented in this paper analyze the application performance based on various benchmark scenarios. In addition, we performed load and scalability testing of an IoT application that adopts the architecture components in a hybrid scenario (real devices and emulated devices). The results obtained were promising. The architecture had a good response to the increase in the workload, not presenting errors, crashes or instabilities due to the increase in IoT data traffic. Moreover, analyzing the overhead generated by the architecture components, there was no performance reduction or service unavailability. Such results point to the fact that the adoption of microservices in the construction of IoT systems can bring effective benefits without jeopardizing their performance due to eventually generated overheads.

Profiling and Optimizing Java Streams

The Stream API was added in Java 8 to allow the declarative expression of data-processing logic, typically map-reduce-like data transformations on collections and datasets. The Stream API introduces two key abstractions. The stream, which is a sequence of elements available in a data source, and the stream pipeline, which contains operations (e.g., map, filter, reduce) that are applied to the elements in the stream upon execution. Streams are getting popular among Java developers as they leverage the conciseness of functional programming and ease the parallelization of data processing. Despite the benefits of streams, in comparison to data processing relying on imperative code, streams can introduce significant overheads which are mainly caused by extra object allocations and reclamations, and the use of virtual method calls. As a result, developers need means to study the runtime behavior of streams in the goal of both mitigating such abstraction overheads and optimizing stream processing. Unfortunately, there is a lack of dedicated tools able to dynamically analyze streams to help developers specifically locate issues degrading application performance. In this paper, we address the profiling and optimization of streams. We present a novel profiling technique for measuring the computations performed by a stream in terms of elapsed reference cycles, which we use to locate problematic streams with a major impact on application performance. While accuracy is crucial to this end, the inserted instrumentation code causes the execution of extra cycles, which are partially included in the profiles. To mitigate this issue, we estimate and compensate for the extra cycles caused by the inserted instrumentation code. We implement our approach in StreamProf that, to the best of our knowledge, is the first dedicated stream profiler for the Java Virtual Machine (JVM). With StreamProf, we find that cycle profiling is effective to detect problematic streams whose optimization can enable significant performance gains. We also find that the accurate profiling of tasks supporting parallel stream processing allows the diagnosis of load imbalance according to the distribution of stream-related cycles at a thread level. We conduct an evaluation on sequential and parallel stream-based workloads that are publicly available in three different sources. The evaluation shows that our profiling technique is efficient and yields accurate profiles. Moreover, we show the actionability of our profiles by guiding stream-related optimizations on two workloads from Renaissance. Our optimizations require the modification of only a few lines of code while achieving speedups up to a factor of 5x. Java streams have been extensively studied by recent work, focusing on both how developers are using streams and how to optimize them. Current approaches in the optimization of streams mainly rely on static analysis techniques that overlook runtime information, suffer from important limitations to detect all streams executed by a Java application, or are not suitable for the analysis of parallel streams. Understanding the dynamic behavior of both sequential and parallel stream processing and its impact on application performance is crucial to help users make better decisions while using streams.

Multi-source domain generalization and adaptation toward cross-subject myoelectric pattern recognition

Objective. Myoelectric pattern recognition (MPR) has shown satisfactory performance under ideal laboratory conditions. Nevertheless, the individual variances lead to dramatic performance degradation in cross-user MPR applications. It is crucial to enable the myoelectric interface to adapt to multiple users’ surface electromyography (sEMG) distributions in practical. Approach. Domain adaptation (DA) is a promising approach to tackle cross-user challenges due to its ability to diminish the divergence between individual users’ EMG distributions and escalate model generalization performance. However, existing DA methods in sEMG control are based on single-source domain adaptation (SDA). SDA solely mixes multiple training users’ data as a combined source domain and attempts to align with a novel user. This simple data mixing manner ignores the sEMG distribution variations between disparate training users, leading to an insufficient variance elimination and lower performance. To this end, this paper proposes a multi-source synchronize domain adaptation framework with both DA and domain generalization (DG) capability. This multi-source framework aligns each source user and the new user in individual feature spaces, which better transfers the knowledge of existing users to the new user. Moreover, we retain the source-combined data to preserve the effectiveness of SDA. The property was further confirmed by evaluating the performance of the proposed method on data from nine subjects performing six tasks. Main results. Experiment results prove that the proposed multi-source framework achieved both positive DG and DA performance in a cross-user classification manner. Significance. This work demonstrates the usability and feasibility of the proposed multi-source framework in cross-user myoelectric control.

Painless and accurate medical image analysis using deep reinforcement learning with task-oriented homogenized automatic pre-processing

Pre-processing is widely applied in medical image analysis to remove the interference information. However, the existing pre-processing solutions mainly encounter two problems: (i) it is heavily relied on the assistance of clinical experts, making it hard for intelligent CAD systems to deploy quickly; (ii) due to the personnel and information barriers, it is difficult for medical institutions to conduct the same pre-processing operations, making a deep model that performs well on a specific medical institution difficult to achieve similar performances on the same task in other medical institutions. To overcome these problems, we propose a deep-reinforcement-learning-based task-oriented homogenized automatic pre-processing (DRL-HAPre) framework to overcome these two problems. This framework utilizes deep reinforcement learning techniques to learn a policy network to automatically and adaptively select the optimal pre-processing operations for the input medical images according to different analysis tasks, thus helping the intelligent CAD system to achieve a rapid deployment (i.e., painless) and maintain a satisfactory performance (i.e., accurate) among different medical institutes. To verify the effectiveness and advantages of the proposed DRL-HAPre framework, we further develop a homogenized automatic pre-processing model based on the DRL-HAPre framework to realize the automatic pre-processing of key region selection (called HAPre-KRS) in the pneumonia image classification task. Extensive experimental studies are conducted on three pediatric pneumonia classification datasets with different image qualities, and the results show that: (i) There does exist a hard-to-reproduce problem in clinical practices and the fact that having different medical image qualities in different medical institutes is an important reason for the existing of hard-to-reproduce problem, so it is compelling to propose homogenized automatic pre-processing method. (ii) The proposed HAPre-KRS model and DRL-HAPre framework greatly outperform three kinds of state-of-the-art baselines (i.e., pre-processing, attention and pneumonia baseline), and the lower the medical image quality, the greater the improvements of using our HAPre-KRS model and DRL-HAPre framework. (iii) With the help of homogenized pre-processing, HAPre-KRS (and DRL-HAPre framework) can greatly avoid performance degradation in real-world cross-source applications (i.e., thus overcoming the hard-to-reproduce problem).

Data-Driven Application-Oriented Reliability Model of a High-Performance Computing System

Reliability analysis and performance evaluation are complementary methods to quantify nonfunctional aspects of a system. However, a range of factors such as concurrency and heterogeneity quickly exacerbate the state-space explosion problem when attempting detailed system-level modeling and simulation of high-performance computing (HPC) systems. To overcome these impediments to modeling and analysis, this article develops a hierarchical model of an application that implements checkpointing running in an HPC environment subject to application, network, and system-wide outages. The modeling approach ensures that the number of states is linear in the number of checkpoints and possesses a low constant factor for the number of recovery states most relevant to the external influences contributing to degraded application performance. We illustrate the types of analysis enabled by the model through a series of examples with parameters determined empirically from data logs of the Blue Waters supercomputer located at the University of Illinois at Urbana–Champaign. A comprehensive comparative analysis of the model parameters indicates that lowering the failure rate of network nodes would most significantly reduce application downtime. We also discuss how the modeling approach can be used to objectively assess both current and hypothetical future systems to identify competitive designs and enhancements.

Methodology for Power-Performance Trade-Off Management in Real-Time Embedded Applications

An increasing number of battery-powered devices that are used outdoors or in mobile systems put emphasis on the power and energy efficiency as a form of trade-off between application performance and system power consumption. However, lack of objective metrics for the evaluation of application performance degradation poses difficulties for managing such trade-offs in real-time applications. The proposed methodology introduces metrics for modeling of application performance and the technique for its control, enabling more efficient power–performance trade-off management. The methodology allows for selective system performance degradation and fine-grained control of system behavior in the power–performance domain by extending the set of operating point parameters controllable through real-time application. The utilization and the effectiveness of the proposed methodology is evaluated in a simulated environment for different scenarios of the application execution, including system operation above the utilization bounds.

Improving Bank-Level Parallelism for In-Memory Checkpointing in Hybrid Memory Systems

Checkpoint/recovery has been widely used in many high available and reliable systems. This paper proposes Shadow, an application-transparent and in-memory checkpointing mechanism based on hybrid memory system composed of DRAM and emerging Non-Volatile Memory (NVM). Shadow adopts a pre-copy based checkpointing mechanism to reduce the system downtime. It supports fine-grained and incremental checkpointing at frequencies up to 100 times per second. Under this context, the checkpointing can significantly degrade application performance due to memory contention between applications and the checkpointing process. Previous checkpointing mechanisms on hybrid memory systems have focused on the performance of checkpointing, and have overlooked the impact of memory contention on the application performance. In this paper, we mitigate the memory contention at the bank level by carefully scheduling memory requests to fully leverage the idle time slots of different memory banks. Moreover, if bank conflicts are unavoidable, Shadow promotes the priority of applications' memory requests to lessen their access latencies. By redesigning the memory controllers of DRAM and NVM, we implement a hardware-assisted checkpointing mechanism that can directly transfer data from working memory to the checkpoint in NVM, without any intervention of CPUs. Our evaluation shows that Shadow can reduce memory bank conflicts between applications and checkpointing by 75%, and decrease applications' memory read request latency by 28% on average compared to the pre-copy based checkpointing. Moreover, Shadow can also reduce checkpointing overhead by 42% and 16% on average compared to the stop-and-copy and pre-copy based checkpointing approaches, respectively.