Shahram Latifi

Sep: A Fixed Degree Regular Network for MassivelyParallel Systems

We propose a family of regular Cayley network graphs of degree three based on permutation groups for design of massively parallel systems. These graphs are shown to be based on the shuffle exchange operations, to have logarithmic diameter in the number of vertices, and to be maximally fault tolerant. We investigate different algebraic properties of these networks (including fault tolerance) and propose a simple routing algorithm. These graphs are shown to be able to efficiently simulate or embed other permutation group based graphs; thus they seem to be very attractive for VLSI implementation and for applications requiring bounded number of I/O ports as well as to run existing applications for other permutation group based network architectures.

Images similarity estimation by processing compressed data

Abstract A method is proposed to evaluate the similarity of images compressed by a given digital wavelet transform. The method allows for comparing lossless- or lossy-compressed images. Two features that describe the image structural content, namely edge point locations and edge density, are computed directly from multiscale data. The distance between two images is computed in one- or two-dimensional space, depending on the image type and the choice of features selected for processing. The method facilitates content-based image querying and automatic database retrieval. Additionally, images can be sorted and appropriately indexed with respect to such characteristics as smoothness, texture direction, repetition period and the average number of edge points in spatial period.

Variable-Dilation Embeddings of Hypercubes into Star Graphs: Performance Metrics, Mapping Functions, and Routing

We present load 1 embeddings of a k-dimensional hypercube Q k into an n-dimensional star graph S n . Dimension i links of Q k are mapped into paths of length at most d i in S n , where d i varies with i rather than being fixed. Our embeddings are an attractive alternative to previously known techniques, producing small average dilation and small average congestion without sacrificing expansion. We provide a thorough characterization of our embeddings, which spans several combinations of node mapping functions and routing algorithms in S n .

A wavelet-based technique for image similarity estimation

In this paper we proposed a method to evaluate the similarity of images compressed by a given digital wavelet transform, which allows for comparing lossless or lossy-compressed images. Two features that describe the image structural content, edge point locations and edge density, are computed directly from multiscale data. Depending on the image type and the feature selections for processing, the distance between two images is computed in one- or two-dimensional space. This method facilitates content-based image querying and automatic database retrieval. In addition, images can be sorted and appropriately indexed with respect to such global characteristics as smoothness, texture direction and repetition period.

Modern flash technologies: a flash translation layer perspective

Since the introduction of the first flash memory in 1984, flash memory has been a very important member of the non-volatile semiconductor memory family due to its advantages, such as high density, low-cost, shock resistance, fast access time, low-power consumption and reliability. Despite the advantages, flash memory is still facing many technical limitations that need to be further studied. Various solutions have been developed to improve the performance of flash memory by overcoming the technical limitations; however, flash translation layer has a great impact on the overall performance improvement due to its cost-efficiency and usefulness. In this paper, three main aspects of flash memory are discussed from a perspective of flash translation layer. First, we discuss and classify flash translation layer based on the mapping method. Second, we discuss the current technical limitations that flash memory is facing and how these can be overcome by adopting flash translation layer. Third, we investigate the latest flash translation layer schemes that have been recently studied and proposed, and analyse their advantages and drawbacks.

Future prospects of DRAM: emerging alternatives

DRAM has been the dominant memory for todays computing memory systems due to its high data rate up to 1,866 Mbps. Researchers and developers of DRAM have been striving to improve the performance in terms of higher speed and higher density with lower bit cost. The conventional method of reducing the cost is to scale down. However, scaling-down is becoming a very difficult process because the scale has already reached a 20 nm node which has been recently developed by Samsung. As it becomes difficult to scale down, it becomes harder to have a higher density and yield, and of course, the cost cannot be reduced. Therefore, various memories have been studied and developed as potential candidates for the next generation DRAM. In this paper, we investigate the technical limitations that conventional DRAM needs to overcome, especially in terms of scaling-down. Then, we discuss the emerging memories that can be used as alternatives to DRAM and we point out what advantages they have.