An-Chow Lai

Dead-block prediction & dead-block correlating prefetchers

Effective data prefetching requires accurate mechanisms to predict both “which” cache blocks to prefetch and “when” to prefetch them. This paper proposes the Dead-Block Predictors (DBPs), trace-based predictors that accurately identify “when” an Ll data cache block becomes evictable or “dead”. Predicting a dead block significantly enhances prefetching lookahead and opportunity, and enables placing data directly into Ll, obviating the need for auxiliary prefetch buffers. This paper also proposes Dead-Block Correlating Prefetchers (DBCPs), that use address correlation to predict “which” subsequent block to prefetch when a block becomes evictable. A DBCP enables effective data prefetching in a wide spectrum of pointer-intensive, integer, and floating-point applications. We use cycle-accurate simulation of an out-of-order superscalar processor and memory-intensive benchmarks to show that: (1) dead-block prediction enhances prefetching lookahead at least by an order of magnitude as compared to previous techniques, (2) a DBP can predict dead blocks on average with a coverage of 90% only mispredicting 4% of the time, (3) a DBCP offers an address prediction coverage of 86% only mispredicting 3% of the time, and (4) DBCPs improve performance by 62% on average and 282% at best in the benchmarks we studied.

Selective, accurate, and timely self-invalidation using last-touch prediction

Communication in cache-coherent distributed shared memory (DSM) often requires invalidating (or writing back) cached copies of a memory block, incurring high overheads. This paper proposes Last-Touch Predictors (LTPs) that learn and predict the “last touch” to a memory block by one processor before the block is accessed and subsequently invalidated by another. By predicting a last-touch and (self-)invalidating the block in advance, an LTP hides the invalidation time, significantly reducing the coherence overhead. The key behind accurate last-touch prediction is trace-based correlation, associating a last-touch with the sequence of instructions (i.e., a trace) touching the block from a coherence miss until the block is invalidated. Correlating instructions enables an LTP to identify a last-touch to a memory block uniquely throughout an applications execution. In this paper, we use results from running shared-memory applications on a simulated DSM to evaluate LTPs. The results indicate that: (1) our base case LTP design, maintaining trace signatures on a per-block basis, substantially improves prediction accuracy over previous self-invalidation schemes to an average of 79%; (2) our alternative LTP design, maintaining a global trace signature table, reduces storage overhead but only achieves an average accuracy of 58%; (3) last-touch prediction based on a single instruction only achieves an average accuracy of 41% due to instruction reuse within and across computation; and (4) LTP enables selective, accurate, and timely self-invalidation in DSM, speeding up program execution on average by 11%.

Memory sharing predictor: the key to a speculative coherent DSM

Recent research advocates using general message predictors to learn and predict the coherence activity in distributed shared memory (DSM). By accurately predicting a message and timely invoking the necessary coherence actions, a DSM can hide much of the remote access latency. This paper proposes the Memory Sharing Predictors (MSPs), pattern-based predictors that significantly improve prediction accuracy and implementation cost over general message predictors. An MSP is based on the key observation that to hide the remote access latency, a predictor must accurately predict only the remote memory accesses (i.e., request messages) and not the subsequent coherence messages invoked by an access. Simulation results indicate that MSPs improve prediction accuracy over general message predictors from 81% to 93% while requiring less storage overhead.This paper also presents the first design and evaluation for a speculative coherent DSM using pattern-based predictors. We identify simple techniques and mechanisms to trigger prediction timely and perform speculation for remote read accesses. Our speculation hardware readily works with a conventional full-map write-invalidate coherence protocol without any modifications. Simulation results indicate that performing speculative read requests alone reduces execution times by 12% in our shared-memory applications.

Cohesion: an efficient distributed shared memory system supporting multiple memory consistency models

This paper describes a prototype of DSM called Cohesion which supports two memory consistency models, namely Sequential consistency and Release consistency, within a single program to improve the performance and supports wide-variety of parallel programs for the system. Memory that is sequentially consistent is further divided into object-based and conventional (page-based) memory; where they are constructed in user-level and kernel-level, respectively. In object-based memory, the shared data are kept consistent at the granularity of an object; it is provided to improve the performance of the fine-grained parallel applications that may incur a significant overhead in conventional or release memory, as well as to eliminate unnecessary movement of the pages which are protected in a critical section. On the other hand, the Release consistency model is supported in Cohesion to attack the problem of excessive network traffic and false sharing. Cohesion programs are written in C++, and the annotation of shared objects for release and object-based memory is accomplished by inheriting a system-provided base class. Finally, three application programs including Matrix Multiplication, SOR, and Nbody have been employed to evaluate the efficiency of Cohesion. In addition, a Producer-Consumer program is tested to show that the object-based memory will benefit us in a critical section.<<ETX>>

Load balancing in distributed shared memory systems

Despite the fast evolution of Distributed Shared Memory (DSM) systems, the load balance problem has not received enough attentions. This problem arises naturally after multithreading was introduced to DSM systems few years ago. The cognizance of it would bring us a significant improvement in system performance. In this paper, we address it by proposing and experimentally evaluating a load balancing method called Dependence-Driven Load Balancing (DDLB) that is dedicated for multithreaded DSM systems. The most attractive feature of this method is to take thread dependence into account in making decisions for migration. In contrast to existing thread scheduling works which for the most part have relied on the information supplied before execution, we require no a priori information. Typically, DDLB embraces three policies, i.e. transfer policy, location policy and selection policy, and applies affinity scheduling. Finally, from the experimental results, the performance of the system with load balancing is improving evidently.

Data privacy preserving scheme using generalised linear models

Abstract When releasing data for public use, statistical agencies seek to reduce the risk of disclosure, while preserving the utility of the release data. Commonly used approaches (such as adding random noises, top coding variables and swapping data values) will distort the relationships in the original data. To preserve the utility and reduce the risk of disclosure for the released data, we consider the synthetic data approach in this paper where we release multiply imputed partially synthetic data sets comprising original data values, and with values at high disclosure risk being replaced by synthetic values. To generate such synthetic data, we introduce a new variant of factored regression model proposed by Lee and Mitra in 2016. In addition, we take a step forward to propose a new algorithm in identifying the original data that need to be replaced with synthetic data. More importantly, the algorithm that can identify the original data with high disclosure risk can be applied on other existing statistical disclosure control schemes. By using our proposed scheme, data privacy can be preserved since it is difficult to identify the individual under the scenario that the released synthetic data are not entirely similar with the original data. Besides, valid inference about the data can be made using simple combining rules, which take the uncertainty due to the presence of synthetic values. To evaluate the performance of our proposed scheme in terms of the risk of disclosure and the utility of the released synthetic data, we conduct an experiment on a data set taken from 1987 National Indonesia Contraceptive Prevalence. The results justify the applicability of our proposed data privacy preserving scheme in reducing the risk of disclosure while preserving the utility of the released data.

Statistical Disclosure Control for Data Privacy Using Sequence of Generalised Linear Models

When releasing data for public use, statistical agencies seek to reduce the risk of disclosure, while preserving the utility of the release data. Common approaches such as adding random noises, top coding variables and swapping data values will distort the relationships in the original data. To achieve the aforementioned properties, we consider the synthetic data approach in this paper where we release multiply imputed partially synthetic data sets comprising original data values, and with values at high disclosure risk being replaced by synthetic values. To generate such synthetic data, we introduce a new variant of factored regression model proposed by Lee and Mitra in 2016. In addition, we take a step forward to propose a new algorithm in identifying the original data that need to be replaced with synthetic data. By using our proposed methods, data privacy can be preserved since it is difficult to identify the individual under the scenario that the released synthetic data are not entirely similar with the original data. Besides, valid inference about the data can be made using simple combining rules, which take the uncertainty due to the presence of synthetic values. To evaluate the performance of our proposed methods in term of the risk of disclosure and the utility of the released synthetic data, we conduct an experiment on a dataset taken from 1987 National Indonesia Contraceptive Prevalence.

A generic sun-tracking algorithm for on-axis solar collector in mobile platforms

This paper proposes a novel dynamic sun-tracking algorithm which allows accurate tracking of the sun for both non-concentrated and concentrated photovoltaic systems located on mobile platforms to maximize solar energy extraction. The proposed algorithm takes not only the date, time, and geographical information, but also the dynamic changes of coordinates of the mobile platforms into account to calculate the sun position angle relative to ideal azimuth-elevation axes in real time using general sun-tracking formulas derived by Chong and Wong. The algorithm acquires data from open-loop sensors, i.e. global position system (GPS) and digital compass, which are readily available in many off-the-shelf portable gadgets, such as smart phone, to instantly capture the dynamic changes of coordinates of mobile platforms. Our experiments found that a highly accurate GPS is not necessary as the coordinate changes of practical mobile platforms are not fast enough to produce significant differences in the calculation of ...

Dense-array concentrator photovoltaic system using non-imaging dish concentrator and crossed compound parabolic concentrator

Solar concentrating device plays an important role by making use of optical technology in the design, which can be either reflector or lens to deliver high flux of sunlight onto the Concentrator Photovoltaic (CPV) module receiver ranging from hundreds to thousand suns. To be more competitive compared with fossil fuel, the current CPV systems using Fresnel lens and Parabolic dish as solar concentrator that are widely deployed in United States, Australia and Europe are facing great challenge to produce uniformly focused sunlight on the solar cells as to reduce the cost of electrical power generation. The concept of non-imaging optics is not new, but it has not fully explored by the researchers over the world especially in solving the problem of high concentration solar energy, which application is only limited to be a secondary focusing device or low concentration device using Compound Parabolic Concentrator. With the current advancement in the computer processing power, we has successfully invented the non...

Hybrid face detection with skin segmentation and edge detection

Face detection for low quality images and different face positions is a very challenging task. This paper presents a hybrid method for face detection to these problems. The algorithm starts with image resizing process followed by the Gaussian Mixture Model to calculate the skin likelihood value of pixel in an image. Then, the skin regions are extracted from the background with a proper threshold value obtained adaptively based on image information. This study developed an algorithm to performed face location and detection. This study used face database from University of Ljubljana (Slovenia) Computer Vision Laboratory (CVL), which contains seven 2D images corresponding to 114 different individuals, to evaluate the proposed system. The resolution of the images is 640*480 pixels. Another database, the Bao database which consists of 157 images with image resolutions within 57×85 pixels and 300 × 300 pixels is chosen. The detection accuracy for frontal face and side face images on CVL database is 94.4% and 84.7% respectively. The detection accuracy on Bao database is 93.6%.