Memory Scalability Research Articles

Cr metal thin film memory

As state of the art flash memory technologies scale down to sub 30 nm node, conventional floating gate flash memory approaches its physical scaling limit mainly because of the high gate coupling ratio (GCR) requirement to secure proper memory window. Here, we report a novel flash memory device called Cr metal thin film memory (MTFM) that can circumvent the GCR issue and extend flash memory scalability by employing Cr thin film as a storage layer. Cr metal thin film memory devices with simple and low temperature processes produced a wide memory window of 10 V at the ±18 V voltage sweep with GCR of only 0.3. Such a large window can be adopted for multi-level cell operations, which can further increase the memory density. Also, retention measurement shows more than 10 years retention time due to higher energy barrier between Cr metal and tunnel oxide than conventional poly silicon and tunnel oxide. Cross section transmission electron microscope (TEM) images showed the structure and accurate dimensions of the Cr MTFM device with continuous Cr film and sharp interfaces. As for material characterizations, an amorphous like Cr phase was observed through TEM and x-ray diffraction (XRD). X-ray photoelectron spectroscopy (XPS) confirmed the Cr-Cr bond and Cr-O bond near the Cr surface after evaporation and rapid thermal annealing. This metal thin film memory may open a new route to achieve the terabit level flash memory.

Comparisons of different codes for galacticN-body simulations

Context. N-body simulations are widely used for galactic dynamics studies. Several different algorithms have been developed and a number of authors have made their codes public. Aims. We wish to help potential users of these codes, particularly less experienced ones, to select the appropriate code by giving them relevant information. Methods. We consider different implementations of three different algorithms, namely the Barnes-Hut tree-code algorithm (implemented both in GADGET-2 and as a NEMO program), the fast multipole method (implemented in the FMB code – Fast Multipole with BLAS) and W. Dehnen’s algorithm (implemented in falcON – force algorithm with complexity O(N)). We compare extensively the timing and memory comparisons of these different implementations and discuss other advantages and disadvantages of these codes. Results. In terms of serial executions, falcON is clearly the fastest code, in many cases by a wide margin. In addition, it is much less sensitive to concentration than other codes, it conserves linear momentum, its computational costs depends roughly linearly on the number of particles, and it allows for manipulators. GADGET-2 is the second fastest and, being parallel, can outperform falcON if as ufficient number of processors is used, the number of which, in our tests is between three and ten. Its public version includes an SPH representation of the gas component. FMB is faster than the implementation of the Barnes-Hut algorithm in NEMO, as long as the particle distribution is not too concentrated, but remains considerably slower than falcON and GADGET-2. It has, nevertheless, the highest parallel efficiency and memory scalability. Conclusions. We recommend falcON for pure N-body simulations with a relatively restricted number of particles (up to a few million). In cases of a larger number of particles, a number of processors being available, or studies including gas and its physics, we recommend GADGET-2. FMB provides the highest parallel efficiency and memory scalability, but is considerably slower in execution time and is ill-adapted to the concentrated distributions encountered in galactic work.

Cooperative server clustering for a scalable GAS model on petascale cray XT5 systems

Global Address Space (GAS) programming models are attractive because they retain the easy-to-use addressing model that is the characteristic of shared-memory style load and store operations. The scalability of GAS models depends directly on the design and implementation of runtime libraries on the targeted platforms. In this paper, we examine the memory requirement of a popular GAS run-time library, Aggregate Remote Memory Copy Interface (ARMCI) on petascale Cray XT 5 systems. Then we describe a new technique cooperative server clustering that enhances the memory scalability of ARMCI communication servers. In cooperative server clustering, ARMCI servers are organized into clusters, and cooperatively process incoming communication requests among them. A request intervention scheme is also designed to expedite the return of responses to the initiating processes. Our experimental results demonstrate that, with very little impact on ARMCI communication latency and bandwidth, cooperative server clustering is able to significantly reduce the memory requirement of ARMCI communication servers, thereby enabling highly scalable scientific applications. In particular, it dramatically reduces the total execution time of a scientific application, NWChem, by 45% on 2400 processes.

ScaLAPACK's MRRR algorithm

The (sequential) algorithm of Multiple Relatively Robust Representations, MRRR, is a more efficient variant of inverse iteration that does not require reorthogonalization. It solves the eigenproblem of an unreduced symmetric tridiagonal matrix T ∈ R n × n at O ( n 2 ) cost. The computed normalized eigenvectors are numerically orthogonal in the sense that the dot product between different vectors is O ( n ϵ), where ϵ refers to the relative machine precision. This article describes the design of ScaLAPACK's parallel MRRR algorithm. One emphasis is on the critical role of the representation tree in achieving both adequate accuracy and parallel scalability. A second point concerns the favorable properties of this code: subset computation, the use of static memory, and scalability. Unlike ScaLAPACK's Divide & Conquer and QR, MRRR can compute subsets of eigenpairs at reduced cost. And in contrast to inverse iterations which can fail, it is guaranteed to produce a satisfactory answer while maintaining memory scalability. ParEig, the parallel MRRR algorithm for PLAPACK, uses dynamic memory allocation. This is avoided by our code at marginal additional cost. We also use a different representation tree criterion that allows for more accurate computation of the eigenvectors but can make parallelization more difficult.

동적 XML 조각 스트림에 대한 메모리 효율적 질의 처리

ABSTRACT This paper is on query processing in the mobile devices where memory capacity is limited. In case that a query against a large volume of XML data is processed in such a mobile device, techniques of fragmenting the XML data into chunks and of streaming and processing them are required. Such techniques make it possible to process queries without materializing the XML data in its entirety. The previous schemes such as XFrag[4], XFPro[5], XFLab[6] are not scalable with respect to the increase of the size of the XML data because they lack proper memory management capability. After some information on XML fragments necessary for query processing is stored, it is not deleted even after it becomes of no use. As such, when the XML fragments are dynamically generated and infinitely streamed, there could be no guarantee of normal completion of query processing. In this paper, we address scalability of query processing over dynamic XML fragment stream, proposing techniques of deleting information on XML fragments accumulated during query processing in order to extend the previous schemes. The performance experiments through implementation showed that our extended schemes considerably outperformed the previous ones in memory efficiency and scalability with respect to the size of the XML data.Key Words:XML Fragment Stream, XML Query Processing, XML Labelling, Hole-Filler Model

The prospects of non-volatile phase-change RAM

Phase-change read-and-write memory (PRAM) is a promising memory that can solve the problems of conventional memory—scalability, read/write speed, and reliability. We will review the opportunities and technical challenges of PRAM. The most important challenge of PRAM is the reduction of the writing current. Various approaches to reduce the writing current will be reviewed and the prospects of PRAM are discussed.

Perspectives in astrophysical databases

Astrophysics has become a domain extremely rich of scientific data. Data mining tools are needed for information extraction from such large data sets. This asks for an approach to data management emphasizing the efficiency and simplicity of data access; efficiency is obtained using multidimensional access methods and simplicity is achieved by properly handling metadata. Moreover, clustering and classification techniques on large data sets pose additional requirements in terms of computation and memory scalability and interpretability of results. In this study we review some possible solutions.

Extraction of Two-Node Bridges From Large Industrial Circuits

Enumeration and prioritization of highly probable bridges based on the circuit layout and manufacturing defect data is a key step in defect-based testing. Existing solutions either do not scale to large designs or compromise on the accuracy of the computation when applied to very large circuits. This paper presents a scalable and efficient methodology to accurately extract two-node bridges from very large circuits. To our knowledge, this is the first solution to be presented that can process such large industrial designs accurately. It also naturally addresses two important issues viz. through the cell routing and name propagation. Experimental results illustrating key features of the algorithm, including scalability and efficient memory usage, are presented.

Impact of reordering on the memory of a multifrontal solver

This paper is concerned with the memory usage of sparse direct solvers, which depends on the ordering of the unknowns and the scheduling of the computational tasks. We study the influence of state-of-the-art sparse matrix reordering techniques on the memory usage of a multifrontal solver. Concerning the scheduling, the memory usage depends on the tree traversal and how the tasks are assigned to the processors. We analyze the memory scalability when a dynamic scheduling strategy mainly based on the balance of the workload is used. Finally we give hints to improve the parallel memory behaviour.

Fast and accurate text classification via multiple linear discriminant projections

Abstract.Support vector machines (SVMs) have shown superb performance for text classification tasks. They are accurate, robust, and quick to apply to test instances. Their only potential drawback is their training time and memory requirement. For n training instances held in memory, the best-known SVM implementations take time proportional to n a, where a is typically between 1.8 and 2.1. SVMs have been trained on data sets with several thousand instances, but Web directories today contain millions of instances that are valuable for mapping billions of Web pages into Yahoo!-like directories. We present SIMPL, a nearly linear-time classification algorithm that mimics the strengths of SVMs while avoiding the training bottleneck. It uses Fisher's linear discriminant, a classical tool from statistical pattern recognition, to project training instances to a carefully selected low-dimensional subspace before inducing a decision tree on the projected instances. SIMPL uses efficient sequential scans and sorts and is comparable in speed and memory scalability to widely used naive Bayes (NB) classifiers, but it beats NB accuracy decisively. It not only approaches and sometimes exceeds SVM accuracy, but also beats the running time of a popular SVM implementation by orders of magnitude. While describing SIMPL, we make a detailed experimental comparison of SVM-generated discriminants with Fisher's discriminants, and we also report on an analysis of the cache performance of a popular SVM implementation. Our analysis shows that SIMPL has the potential to be the method of choice for practitioners who want the accuracy of SVMs and the simplicity and speed of naive Bayes classifiers.

Optimization techniques for parallel irregular reductions

Different parallelization techniques have been proposed in the literature for irregular reductions in the context of shared memory multiprocessors. They may be classified into two broad families: those based on privatization of the reduction arrays and those based on the partitioning of the reduction arrays. Methods in the first family are simple but no data locality is exploited and their memory scalability is low. On the other hand, methods in the second family are more complex as they require an inspection phase but they exploit data locality and scale up better in memory. Focusing on partitioning-based methods, although they exhibit a good performance in a wide variety of irregular codes, some specific input data patterns may exist for which the performance is lowered. In particular these kind of access patterns may reduce the exploited parallelism by the method or introduce workload unbalances. In order to mitigate these negative effects, we propose three optimizations for a specific partitioning-based method (DWA–LIP). These optimizations try to increase the exploited parallelism, balance the workload and reduce the effect of high contention degree regions in the reduction arrays. Efficient implementations of the proposed optimizations for the DWA–LIP method have been tested experimentally, and compared with other methods for parallelizing irregular reductions.

Parallel Numeric Algorithms On Faster Computers

Parallel Numeric Algorithms On Faster Computers

Dynamic rate scaling of coded digital video for IVOD applications

In this paper, a memory scalable architecture for rate scaling of pre-coded digital video is proposed. In this architecture, the anchor frames are compressed and scaled to fit the size of the memory used as the frame store, and motion compensation is performed in frequency domain. Its advantages include memory saving and memory scalability which make it possible to allocated the memory resource according to the on-line payload of the system in the VOD environment. Therefore, both high efficiency and flexibility can be obtained in respect of the utilization of memory. Another problem addressed in this paper is the rate control of the rate scaling transcoder. As a matter of fact, the content creation encoding and rate scaling can be regarded as a two-pass encoding system. In this paper, a rate control algorithm with look-ahead bit allocation and direct AC DCT bit usage profile tracking is implemented in the rate scaling transcoder based on the results of the first pass encoding.

Superlinear speedup of an efficient sequential algorithm is not possible

Strong superlinear speedup has been discovered in large scale simulations of parallel 3D DEM for complex-shaped particles, which is based on an algorithm of spatial domain decomposition, and exhibits the “high-CPU-low-memory” characteristics. The interpretation of this phenomenon requires a careful examination of the speedup theory and practice in the field of parallel computing. The superlinear speedup is investigated from three perspectives: (i) memory footprint per process, (ii) cache miss rates of L1, L2 and L3 level caches, and (iii) uniprocessor performance, using a wide range of problem size (across five orders of magnitude of simulation scale regarding number of particles) and number of compute nodes (1–2048 nodes) on DoD supercomputers. The Performance-API (PAPI) is employed in the source code to measure cache miss rate and FLOPS. The strong scaling measurements show that cache miss rate is sensitive to the memory consumption shrinkage per processor, and the last level cache (LLC) contributes most significantly to the strong superlinear speedup among all of the three cache levels, and this is also revealed in the weak scaling measurements. The findings are associated with the inherently perfect scalability of 3D DEM: its memory scalability function is a nonlinearly decreasing function of the number of processors. In addition, a constant (non-increasing) uniprocessor FLOPS performance w.r.t problem size can also contribute to the superlinear speedup.The superlinear speedup is a common phenomenon for large scale 3D DEM simulations of complex-shaped particles, and the larger the scale, the stronger is the superlinear speedup. DEM researchers should take advantage of this effect to speedup their parallel simulations.