Humayun Khalid

Estimating neural networks-based algorithm for adaptive cache replacement

In this paper, we propose an adaptive cache replacement scheme based on the estimating type of neural networks (NNs). The statistical prediction property of such NNs is used in our work to develop a neural network based replacement policy which can effectively identify and eliminate inactive cache lines. This would provide larger free space for a cache to retain actively referenced lines. The proposed strategy may, therefore, yield better cache performance as compared to the conventional schemes. Simulation results for a wide spectrum of cache configurations indicate that the estimating neural network based replacement scheme provides significant performance advantage over existing policies.

Ultrasonic transducer characterization by neural networks

Abstract This paper presents neural network-based system for the characterization of ultrasonic transducers. An automated system for characterizing ultrasonic transducers was designed and built. Different characterizing algorithms were applied and their performance was investigated and compared. It was found that artificial neural network (ANN) techniques, in general, provide better classification as compared to the pattern recognition techniques we applied earlier (M.S. Obaidat, J.W. Ekis, IEEE Transactions on Instrumentation and Measurements, 40 (5) (1991) 847–850). The Moody-Darken Radial Basis Function network (MD-RBFN), Learning Vector Quantization (LVQ) with 52 kohonen neurons, and Fuzzy ARTMAP classification network are the neural networks (NNs) that provided us with a classification accuracy of 100%. Several variants of Backpropagation neural network (BPNN) were tested for this application, and the classification results were seen to vary in the range 6.55%–98.35%. The best performing paradigm among several variants of the Modular Neural Network (MNN), Reinforcement Neural Network (RNN), Probabilistic Neural Network (PNN), and Counterpropagation Neural Network (CPNN) produced a classification accuracy of 78.85%, 31.42%, 45.21%, and 79.28%, respectively. The competitive learning (CL) technique provided poor results as compared to the Self-Organizing-Map (SOM) for preclustering.

A new cache replacement scheme based on backpropagation neural networks

In this paper, we present a new neural network-based algorithm, KORA (Khalid ShadOw Replacement Algorithm), that uses backpropagation neural network (BPNN) for the purpose of guiding the line/block replacement decisions in cache. This work is a continuation of our previous research presented in [1]-[3]. The KORA algorithm attempts to approximate the replacement decisions made by the optimal scheme (OPT). The key to our algorithm is to identify and subsequently discard the dead lines in cache memories. This allows our algorithm to provide better cache performance as compared to the conventional LRU (Least Recently Used), MRU (Most Recently Used), and FIFO (First In First Out) replacement policies. Extensive trace-driven simulations were performed for 30 different cache configurations using different SPEC (Standard Performance Evaluation Corp.) programs. Simulation results have shown that KORA can provide substantial improvement in the miss ratio over the conventional algorithms. Our work opens new dimensions for research in the development of new and improved page replacement schemes for virtual memory systems and disk caches.

Performance of the KORA-2 cache replacement scheme

In this paper, we propose a new strategy (KORA-2) for the replacement of lines in cache memories. The algorithm is efficient and easily implementable. It is basically an extension of our previous work presented in [1]-[5]. Key to our algorithm is to identify and discard inactive lines relatively quickly as opposed to the conventional replacement algorithms. Trace-driven simulations were performed for 42 different cache configurations using benchmark programs from SPEC92 (Standard performance Evaluation Corporation) benchmark suites. Simulation results illustrate that our algorithm can provide a peak value of approximately 8.71% improvement in the miss ratio over the best performing conventional algorithm (LRU) for the selected benchmark trace files generated from SPEC programs. This translates to a savings of hundreds of thousands of misses for typical programs referencing well over 100 million addresses.

Validating trace-driven microarchitectural simulations

Benchmark complexity and sluggish performance analysis create an ever-increasing gap between workload size and the speed of analysis. Experimental results with a new data-sampling methodology, however, show promise in closing this gap.

KORA: a new cache replacement scheme

Abstract In this paper, we present a new neural network based cache replacement algorithm called KORA (Khalid Obaidat Replacement Algorithm). The proposed algorithm uses neural network(s) as a kind of directory in order to identify and distinguish between active and inactive cache lines. Such a classification helps to guide the line replacement decisions. This allows our algorithm to provide better cache performance as compared to the conventional LRU (Least Recently Used), MRU (Most Recently Used), and FIFO (First In First Out) replacement policies (Khalid H, Obaidat MS. A novel cache memory controller: algorithm and simulation, Proceedings of the Summer Computer Simulation Conference (SCSC ’95), Canada, July, 1995. p. 767–772; Khalid H, Obaidat MS. Near-optimal cache replacement policy for high performance computer systems, revised version. IEEE Transactions On Computers, 1997 (submitted for publication); Khalid H, Obaidat MS. High performance cache memory replacement schemes. Proceedings of the 1995 IEEE International Conference on Electronics, Circuits, and Systems, December, 1995, p. 1–7). Extensive trace-driven simulations were performed for a wide variety of cache configurations using different SPEC92 (Standard Performance Evaluation Corporation) benchmark programs. Simulation results illustrate that KORA can provide substantial improvement in the miss ratio over the conventional algorithms. Our work opens a new dimension for research in the development of new and improved page replacement schemes for virtual memory systems and disk caches.

Application of Neural Networks to Cache Replacement

Probabilistic Decision-Based Neural Networks (PDBNNs) can be considered as a special form of Gaussian Mixture Models (GMMs) with trainable decision thresholds. This paper provides detailed illustrations to compare the recognition accuracy and decision boundaries of PDBNNs with that of GMMs through two pattern recognition tasks, namely the noisy XOR problem and the classification of two-dimensional vowel data. The paper highlights the strengths of PDBNNs by demonstrating that their thresholding mechanism is very effective in detecting data not belonging to any known classes. The original PDBNNs use elliptical basis functions with diagonal covariance matrices, which may be inappropriate for modelling feature vectors with correlated components. This paper overcomes this limitation by using full covariance matrices, and showing that the matrices are effective in characterising non-spherical clusters.

Tracing multimedia benchmarks with five degrees of validation

In this paper we have extended our methodology, presented earlier in [1], for generating and validating representative traces. Our novel technique was applied to a relatively realistic and difficult multimedia benchmark suite called MediaMark. We have also introduced a new metric called the K-metric (Khalid - metric) that was used for validation. The aim of our present research was to demonstrate that the proposed methodology can be successful even for complex and challenging benchmarks like multimedia benchmarks. Earlier our methodology was shown to be the most successful one as compared to the popular contemporary techniques for tracing relatively simple and primitive suite of applications contained within SPEC™95 benchmark suite [1]. Experimental results in this article demonstrate that our methodology works in the worst case scenarios.

Simulation study of a novel cache replacement algorithm

This work proposes a new scheme for the replacement of cache lines in computer systems. Performance of the proposed algorithm was tested by conducting simulation experiments. Several simulation models were developed for the cache and neural network paradigms. Our simulation engine was driven by traces from the real world workloads/benchmarks. The proposed strategy uses learning properties of non- estimating type of neural networks to understand the replacement phenomenon and guide the replacement decisions made by the cache controller. Therefore, the strategy was successful in being able to eliminate dead lines from the cache memory more efficiently as compared to the conventional algorithms. We observed from the simulation experiments that a well- designed non-estimating neural network- based replacement policy does provide excellent performance as compared to the overwhelmingly used LRU scheme. The new approach can be applied to the page replacement and prefetching algorithms in virtual memory systems.

Performance evaluation of multimedia systems with MPEG-2 bitstreams

MPEG-2 has become a compression standard for multimedia applications. Performance analysts, designers, and researchers have been using mpeg-2 bitstreams for evaluating the multimedia performance of their systems. There exist a significant difference in the characteristics and behavior of such bitstreams. The bitstream set, used for performance evaluation, that does not demonstrate such variations results in erroneous evaluations and projections. In this paper, we have proposed a methodology for improving the performance estimations/projections with mpeg-2 data. Our simulation results demonstrate variations in the mpeg-2 bitstreams and the consequence it can have when data from only one of the several classes is chosen for system measurement and evaluation