Perceptron learning for reuse prediction

Recent years have seen a large volume of proposals on managing the shared last-level cache (LLC) of chip-multiprocessors (CMPs). However, most of these proposals primarily focus on reducing the amount of destructive interference between competing independent threads of multi-programmed workloads. While very few of these studies evaluate the proposed policies on shared memory multi-threaded applications, they do not improve constructive cross-thread sharing of data in the LLC In this paper, we characterize a set of multi-threaded applications drawn from the PARSEC, SPEC OMP, and SPLASH-2 suites with the goal of introducing sharing-awareness in LLC replacement policies. We motivate our characterization study by quantifying the potential contributions of the shared and the private blocks toward the overall volume of the LLC hits in these applications and show that the shared blocks are more important than the private blocks. Next, we characterize the amount of sharing-awareness enjoyed by recent proposals compared to the optimal policy. We design and evaluate a generic oracle that can be used in conjunction with any existing policy to quantify the potential improvement that can come from introducing sharing-awareness. The oracle analysis shows that introducing sharing-awareness reduces the number of LLC misses incurred by the least-recently-used (LRU) policy by 6% and 10% on average for a 4MB and 8MB LLC respectively. A realistic implementation of this oracle requires the LLC controller to have the capability to accurately predict, at the time a block is filled into the LLC, whether the block will be shared during its residency in the LLC. We explore the feasibility of designing such a predictor based on the address of the fill and the program counter of the instruction that triggers the fill. Our sharing behavior predictability study of two history-based fill-time predictors that use block addresses and program counters concludes that achieving acceptable levels of accuracy with such predictors will require other architectural and/or high-level program semantic features that have strong correlations with active sharing phases of the LLC blocks.

Closely coupled CPU and GPGPU system with the shared last level cache (LLC) enables fine-grained data exchange. However, traditional data exchange causes unnecessary LLC misses and degrades the entire system performance. We believe that the cache organization is not suitable for the temporary data exchange in the closely coupled system. We analyze the memory access pattern and discover the inefficiency data exchange. When the exchanged data cannot fit in the LLC, the low LLC hit rate exacerbates core stalls and memory contention. We also show that the stalls cannot be entirely covered by increasing the compute load or parallelism. In previous work, a simple LLC buffer is introduced to replace the cache with an architecture-supported data queue. However, the simple design limits the data element size and requires a potentially very large storage for pending requests. In this paper, we propose an improved LLC buffer. It adopts element-atom data organization to enable data exchange of arbitrary size. A simple hardware-software collaborated protocol is adopted, and eliminates the pending requests. The experiment results reveal that it has an average speedup of 48.2% compared with the traditional way, but incurs a 7.5% slowdown compared with the simple LLC buffer due to the overhead of the protocol. We also compare it with the fine-grain task, which implements a data exchange channel between CPU and GPGPU. The results show that the improved LLC buffer has less storage overhead but higher access efficiency than the fine-grain task.

The size of the Last Level Caches (LLC) in multi-core architectures is increasing, and so is their power consumption. However, most of this power is wasted on unused or invalid cache lines. For dirty cache lines, the LLC waits until the line is evicted to be written back to memory. Hence, dirty lines compete for the memory bandwidth with read requests (prefetch and demand), increasing pressure on the memory controller. This paper proposes a Dead Line and Early Write-Back Predictor (DEWP) to improve the energy efficiency of the LLC. DEWP early evicts dead cache lines with an average accuracy of 94%, and only 2% false positives. DEWP also allows scheduling of dirty lines for early eviction, allowing earlier write-backs. Using DEWP over a set of single and multi-threaded benchmarks, we obtain an average of 61% static energy savings, while maintaining the performance, for both inclusive and non-inclusive LLCs.

Last-Level Cache (LLC) plays an important role in Chip Multi-Processor (CMP). The objective of this work is to optimize the structure and management strategy of LLC. Based on 8-core CMP, a LLC structure based on grouped cores is proposed, where 8 cores are divided into 4 groups. All LLC resources are classified into three types, which are fixed private cache, dynamic private cache and dynamic shared cache. The layout of the LLC structure and the corresponding dynamic partitioning strategy are designed to achieve low access latency and high efficiency. Experimental results on full-system simulator suggest that the proposed structure and method are able to reduce the access latency by 2% to 12% compared with previous works, such as tiled structure, cache-centered structure and core-centered structure. Consequently, performance measured by IPC is improved up to 7%. The contribution of this paper is useful for CMP performance, and applies to not only 8-core CMP but also all small-scale CMPs.

Last-level caches (LLCs) bridge the processor/memory speed gap and reduce energy consumed per access. Unfortunately, LLCs are poorly utilized because of the relatively large occurrence of dead blocks. We propose RADAR, a hybrid static/dynamic dead-block management technique that can accurately predict and evict dead blocks in LLCs. RADAR does dead-block prediction and eviction at the granularity of address regions supported in many of today's task-parallel programming models. The runtime system utilizes static control-flow information about future region accesses in conjunction with past region access patterns to make accurate predictions about dead regions. The runtime system instructs the cache to demote and eventually evict blocks belonging to such dead regions. This paper considers three RADAR schemes to predict dead regions: a scheme that uses control-flow information provided by the programming model (Look-ahead), a history-based scheme (Look-back) and a combined scheme (Look-ahead and Look-back). Our evaluation shows that, on average, all RADAR schemes outperform state-of-the-art hardware dead-block prediction techniques, whereas the combined scheme always performs best.

Conventional translation look-aside buffers(TLBs) are required to complete address translation withshort latencies, as the address translation is on the criticalpath of all memory accesses even for L1 cache hits. Such strictTLB latency restrictions limit the TLB capacity, as the latencyincrease with large TLBs may lower the overall performanceeven with potential TLB miss reductions. Furthermore, TLBsconsume a significant amount of energy as they are accessedfor every instruction fetch and data access. To avoid thelatency restriction and reduce the energy consumption, virtualcaching techniques have been proposed to defer translation toafter L1 cache misses. However, an efficient solution for thesynonym problem has been a critical issue hindering the wideadoption of virtual caching.Based on the virtual caching concept, this study proposes ahybrid virtual memory architecture extending virtual cachingto the entire cache hierarchy, aiming to improve both performanceand energy consumption. The hybrid virtual cachinguses virtual addresses augmented with address space identifiers(ASID) in the cache hierarchy for common non-synonymaddresses. For such non-synonyms, the address translationoccurs only after last-level cache (LLC) misses. For uncommonsynonym addresses, the addresses are translated to physicaladdresses with conventional TLBs before L1 cache accesses. Tosupport such hybrid translation, we propose an efficient synonymdetection mechanism based on Bloom filters which canidentify synonym candidates with few false positives. For largememory applications, delayed translation alone cannot solvethe address translation problem, as fixed-granularity delayedTLBs may not scale with the increasing memory requirements.To mitigate the translation scalability problem, this studyproposes a delayed many segment translation designed for thehybrid virtual caching. The experimental results show that ourapproach effectively lowers accesses to the TLBs, leading tosignificant power savings. In addition, the approach providesperformance improvement with scalable delayed translationwith variable length segments.

In this paper, we consider last level cache (LLC) sharing problems in large-scale cloud networks with a fair payoff function. We formulate the problem as a strategic decision-making problem (i.e., a game). We examine the resource-sharing game with finite and infinite number of players. Exploiting the aggregate structure of the payoff functions, we show that the resource-sharing game has a Nash equilibrium in a wide range of return index. We show that the Nash equilibrium is not an evolutionarily stable strategy in the finite regime. Then, we introduce a myopic mean-field response where each player implements a mean-field-taking strategy. We show that such a mean-field-taking strategy is an evolutionarily stable strategy in both finite and infinite regime. We provide closed-form expression of the optimal pricing that gives an efficient resource-sharing policy. As the number of active players grows without bound, we show that the equilibrium strategy converges to a mean-field equilibrium, and the optimal prices for resources converge to the optimal price of the mean-field game. Then, we address the demand satisfaction problem for which a necessary and sufficient condition for satisfactory solutions is provided. In addition, a very fast mean-field learning algorithm is provided.

The influence of fatigue and related factors on automated perimetric testing was evaluated in both eyes of sixteen normal observers and sixteen patients with early-to-moderate visual field loss using a Digilab 750 automated perimeter and a customized test procedure. False positive rate, false negative rate, and detection sensitivity at 5,10,15, and 20 degrees eccentricity were measured in 1.5-rain intervals throughout a 21-min visual field examination. Half of the normal observers and patients with visual field loss were given a briefrest(1.5-min) midway through each visual field exam to determine whether this would reduce fatigue effects. Our findings revealed that patients displayed considerably higher average false positive and false negative rates than normal observers. However, neither the patients nor normal observers demonstrated any consistent changes in false positive rate or false negative rate as a function of testing duration. In contrast, both normal observers and patients showed an average decrease in sensitivity as a function of increasing test duration with the magnitude of the time-dependent sensitivity loss becoming greater with increasing stimulus eccentricity. Patients demonstrated a larger time-dependent sensitivity loss than normal observers, averaging ~4dB at 20 degrees eccentricity. The introduction of a brief pause midway through the test procedure appeared to reduce the time-dependent sensitivity loss for the second half of the test procedure, especially for greater eccentricities.

Mobile operators are widely adopting Network Functions Virtualization (NFV) to get the benefits of virtualization, including ease of deployment, flexibility, and cost savings. NFV allows multiple virtualized Radio Access Networks (vRANs) to run on commodity hardware enabling joint signal processing and efficient interference management. In addition, mobile operators can run general-purpose workloads alongside vRANs to utilize spare system resources in the NFV infrastructure. In such a consolidated scenario, it is necessary to ensure that the Key Performance Indicators (KPIs) of vRAN workloads are always met. But the workload consolidation could cause high variability and unpredictability in the performance of the deployed vRANs due to contentions for shared system resources like CPU cores, Last Level Cache (LLC), etc. In order to address this problem, we present JARS – a joint allocation of radio and system resources for the NFV infrastructure – that dynamically adjusts system resources such as CPU cores and LLC-ways, and radio resources such as Physical Resource Blocks (PRBs) to ensure KPIs for vRANs and improve overall resource utilization by workload consolidation. We profile srsLTE to determine the minimal CPU core and LLC resource requirements to satisfy KPIs during different traffic loads, which is used in the JARS. Experimental studies on a prototype system show that the proposed JARS outperforms a state-of-the-art scheme by 23%.

On multicore systems contention for shared resources occurs when memory-intensive threads are co-scheduled on cores that share parts of the memory hierarchy, such as last-level caches and memory controllers. Previous work investigated how contention could be addressed via scheduling. A contention-aware scheduler separates competing threads onto separate memory hierarchy domains to eliminate resource sharing and, as a consequence, mitigate contention. However, all previous work on contention-aware scheduling assumed that the underlying system is UMA (uniform memory access latencies, single memory controller). Modern multicore systems, however, are NUMA, which means that they feature non-uniform memory access latencies and multiple memory controllers. We discovered that contention management is a lot more difficult on NUMA systems, because the scheduler must not only consider the placement of threads, but also the placement of their memory. This is mostly required to eliminate contention for memory controllers contrary to the popular belief that remote access latency is the dominant concern. In this work we quantify the effects on performance imposed by resource contention and remote access latency. This analysis inspires the design of a contention-aware scheduling algorithm for NUMA systems. This algorithm significantly outperforms a NUMA-unaware algorithm proposed before as well as the default Linux scheduler. We also investigate memory migration strategies, which are the necessary part of the NUMA contention-aware scheduling algorithm. Finally, we propose and evaluate a new contention management algorithm that is priority-aware.

Accurate liver segmentation from computed tomography (CT) images is problematic due to non-uniform density, weak boundaries and because there may be multiple liver tumors that have heterogeneous intensities in region(s) of interest (ROIs). So we propose a generalized energy framework that harnesses the statistical intensity approximation with image data on graphs. Our statistical energy term takes advantage of the mixture-of-mixtures Gaussian model to approximate the probability density distribution of the liver and background to better differentiate between the two. The probability density estimation can be combined with the spatial cohesion of the graph-based Dirichlet integral by using graph calculus. Matrix decomposition and differentiation are used to minimize our proposed energy functional. We tested our approach on 20 public high-contrast CT images with single and multiple liver tumors. Our method had an average dice similarity coefficient (DSC) of 93.75±1.29%, an average false positive (FP) rate of 9.43±3.52% and an average false negative (FN) rate of 3.48±1.48%. Our method outperformed the benchmark graph-based Random Walker algorithm (average DSC=81.97±4.09%, average FP rate 34.10±10.53%, and average FN rate 7.10±4.35%).

Skyrmions racetrack memory (SKM) has been identified as a promising candidate for future on-chip cache. Similar to many other nanoscale technologies, process variations also adversely impact the reliability and performance of SKM cache. In this work, we propose the first holistic solution for employing SKM as last-level caches. We first present a novel SKM cache architecture and a physical-to-logic mapping scheme based on our comprehensive analysis on working mechanism of SKM. We then model the impact of process variations on SKM cache performance. By leveraging the developed model, we propose a process variation aware data management technique to minimize the performance degradation of SKM cache incurred by process variations. Experimental results show that the proposed SKM cache can achieve a geometric mean of 1.28x IPC improvement, 2x density increase, and 23% energy reduction compared to Domain Wall racetrack memory (DWM) under the same area constraint across 15 workloads. In addition, our dynamic data management technique can further improve the system IPC by 25% w.r.t. the worst-case design.

ObjectiveTo determine the performance of a commercially available AI tool for fracture detection when used in children with osteogenesis imperfecta (OI).Materials and methodsAll appendicular and pelvic radiographs from an OI clinic at a single centre from 48 patients were included. Seven radiologists evaluated anonymised images in two rounds, first without, then with AI assistance. Differences in diagnostic accuracy between the rounds were analysed.Results48 patients (mean 12 years) provided 336 images, containing 206 fractures established by consensus opinion of two radiologists. AI produced a per-examination accuracy of 74.8% [95% CI: 65.4%, 82.7%], compared to average radiologist performance at 83.4% [95% CI: 75.2%, 89.8%]. Radiologists using AI assistance improved average radiologist accuracy per examination to 90.7% [95% CI: 83.5%, 95.4%]. AI gave more false negatives than radiologists, with 80 missed fractures versus 41, respectively. Radiologists were more likely (74.6%) to alter their original decision to agree with AI at the per-image level, 82.8% of which led to a correct result, 64.0% of which were changing from a false positive to a true negative.ConclusionDespite inferior standalone performance, AI assistance can still improve radiologist fracture detection in a rare disease paediatric population. Radiologists using AI typically led to more accurate diagnostic outcomes through reduced false positives. Future studies focusing on the real-world application of AI tools in a larger population of children with bone fragility disorders will help better evaluate whether these improvements in accuracy translate into improved patient outcomes.Key PointsQuestionHow well does a commercially available artificial intelligence (AI) tool identify fractures, on appendicular radiographs of children with osteogenesis imperfecta (OI), and can it also improve radiologists’ identification of fractures in this population?FindingsSpecialist human radiologists outperformed the AI fracture detection tool when acting alone; however, their diagnostic performance overall improved with AI assistance.Clinical relevanceAI assistance improves specialist radiologist fracture detection in children with osteogenesis imperfecta, even with AI performance alone inferior to the radiologists acting alone. The reason for this was due to the AI moderating the number of false positives generated by the radiologists.Graphical

In modern IT infrastructure projects, optimizing system performance and security is crucial for ensuring reliability, efficiency, and compliance. This study explores the role of business analytics in improving IT infrastructure performance and mitigating cybersecurity risks. By utilizing real-world datasets rather than literature-based models, this research applies predictive analytics, machine learning algorithms, and business intelligence tools to enhance IT operations. The findings reveal a 26.7% reduction in CPU usage, 25% improvement in memory utilization, and a 29.2% decrease in network latency, demonstrating the effectiveness of data-driven performance optimization. Additionally, cybersecurity risk assessments using machine learning models resulted in a 14% improvement in threat detection accuracy, a 4% false positive rate, and a 75% reduction in compliance breach risks, ensuring better adherence to security frameworks like ISO 27001 and NIST. The integration of business intelligence dashboards (Tableau, Power BI) enables real-time monitoring of IT risks, enhancing decision-making and proactive threat mitigation. This study contributes to the field by providing a scalable, analytics-driven framework for IT performance enhancement and cybersecurity resilience, bridging the gap between operational efficiency and security risk management. Future research should explore advanced AI-driven automation and real-time adaptive security measures to further strengthen IT infrastructure

Accurate detection and segmentation of multiple brain metastases (BMs) on magnetic resonance image remain challenging, particularly for those involving small lesions (longest axis length <3 mm), due to limitations in sensitivity, precision, and feature representation in existing deep learning frameworks. To address this challenge, we develop TLC-nnUNet, a novel integration of two advancements into transformer-enhanced nnU-Net (T-nnUNet) architecture: (1) local region perceptron (LRP), a loss constraint prioritizing small BM detection by up-weighting underrepresented voxels; and (2) contrastive learning pretraining (CLP), a supervised model pre-training strategy to amplify latent-space divergence between BM and non-BM regions, reducing false positives (FPs). The developed TLC-nnUNet is trained and evaluated on a multi-institutional dataset and achieves state-of-the-art performance, with 89.70% sensitivity, 97.34% precision, and a Dice coefficient of 0.92 at patient level. Further ablation studies confirm the synergistic contributions of each component: LRP enhances small BM detection, while CLP refines feature contrast, reducing FPs. Visualization via t-distributed stochastic neighbor embedding underscores CLP's role in disentangling BM and non-BM latent representations. Compared to existing methods, TLC-nnUNet demonstrates consistent accuracy of detection and segmentation cross lesion sizes. This framework holds significant promise for clinical workflows, enabling precise BM detection and segmentation in stereotactic radiosurgery and reducing manual contouring time.