Generating Efficient Data Movement Code for Heterogeneous Architectures with Distributed-Memory

Programming for parallel architectures that do not have a shared address space is extremely difficult due to the need for explicit communication between memories of different compute devices. A heterogeneous system with CPUs and multiple GPUs, or a distributed-memory cluster are examples of such systems. Past works that try to automate data movement for distributed-memory architectures can lead to excessive redundant communication. In this paper, we propose an automatic data movement scheme that minimizes the volume of communication between compute devices in heterogeneous and distributed-memory systems. We show that by partitioning data dependences in a particular non-trivial way, one can generate data movement code that results in the minimum volume for a vast majority of cases. The techniques are applicable to any sequence of affine loop nests and works on top of any choice of loop transformations, parallelization, and computation placement. The data movement code generated minimizes the volume of communication for a particular configuration of these. We use a combination of powerful static analyses relying on the polyhedral compiler framework and lightweight runtime routines they generate, to build a source-to-source transformation tool that automatically generates communication code. We demonstrate that the tool is scalable and leads to substantial gains in efficiency. On a heterogeneous system, the communication volume is reduced by a factor of 11X to 83X over state-of-the-art, translating into a mean execution time speedup of 1.53X. On a distributed-memory cluster, our scheme reduces the communication volume by a factor of 1.4X to 63.5X over state-of-the-art, resulting in a mean speedup of 1.55X. In addition, our scheme yields a mean speedup of 2.19X over hand-optimized UPC codes.

Mycobacterium tuberculosis DinG Is a Structure-specific Helicase That Unwinds G4 DNA IMPLICATIONS FOR TARGETING G4 DNA AS A NOVEL THERAPEUTIC APPROACH

The significance of G-quadruplexes and the helicases that resolve G4 structures in prokaryotes is poorly understood. The Mycobacterium tuberculosis genome is GC-rich and contains >10,000 sequences that have the potential to form G4 structures. In Escherichia coli, RecQ helicase unwinds G4 structures. However, RecQ is absent in M. tuberculosis, and the helicase that participates in G4 resolution in M. tuberculosis is obscure. Here, we show that M. tuberculosis DinG (MtDinG) exhibits high affinity for ssDNA and ssDNA translocation with a 5' -> 3' polarity. Interestingly, MtDinG unwinds overhangs, flap structures, and forked duplexes but fails to unwind linear duplex DNA. Our data with DNase I footprinting provide mechanistic insights and suggest that MtDinG is a 5' -> 3' polarity helicase. Notably, in contrast to E. coli DinG, MtDinG catalyzes unwinding of replication fork and Holliday junction structures. Strikingly, we find that MtDinG resolves intermolecular G4 structures. These data suggest that MtDinG is a multifunctional structure-specific helicase that unwinds model structures of DNA replication, repair, and recombination as well as G4 structures. We finally demonstrate that promoter sequences of M. tuberculosis PE_PGRS2, mce1R, and moeB1 genes contain G4 structures, implying that G4 structures may regulate gene expression in M. tuberculosis. We discuss these data and implicate targeting G4 structures and DinG helicase in M. tuberculosis could be a novel therapeutic strategy for culminating the infection with this pathogen.

Seventeen-Level Inverter Formed by Cascading Flying Capacitor and Floating Capacitor H-Bridges

A multilevel inverter for generating 17 voltage levels using a three-level flying capacitor inverter and cascaded H-bridge modules with floating capacitors has been proposed. Various aspects of the proposed inverter like capacitor voltage balancing have been presented in the present paper. Experimental results are presented to study the performance of the proposed converter. The stability of the capacitor balancing algorithm has been verified both during transients and steady-state operation. All the capacitors in this circuit can be balanced instantaneously by using one of the pole voltage combinations. Another advantage of this topology is its ability to generate all the voltages from a single dc-link power supply which enables back-to-back operation of converter. Also, the proposed inverter can be operated at all load power factors and modulation indices. Additional advantage is, if one of the H-bridges fail, the inverter can still be operated at full load with reduced number of levels. This configuration has very low dv/dt and common-mode voltage variation.

Comparison of performance of solar photovoltaics on dual axis tracker with fixed axis at 13 degrees N latitude

Tracking systems, that continually orient photovoltaic (PV) panels towards the Sun, are expected to increase the power output from the PV panels. Tremendous amount of research is being done and funds are being spent in order to increase the efficiency of PV cells to generate more power. We report the performance of two almost identical PV systems; one at a fixed latitude tilt and the other on a two-axis tracker. We observed that the fixed axis PV panels generated 336.3 kWh, and the dual-axis Sun-tracked PV panels generated 407.2 kWh during August 2012 March 2013. The tracked panels generated 21.2% more electricity than the optimum tilt angle fixed-axis panels. The cost payback calculations indicate that the additional cost of the tracker can be recovered in 450 days.

A comparative study of failure features in aerospace grade unidirectional and bidirectional woven CFRP composite laminates under four-point bend fatigue loads

Damage mechanisms in unidirectional (UD) and bi-directional (BD) woven carbon fiber reinforced polymer (CFRP) laminates subjected to four point flexure, both in static and fatigue loadings, were studied. The damage progression in composites was monitored by observing the slopes of the load vs. deflection data that represent the stiffness of the given specimen geometry over a number of cycles. It was observed that the unidirectional composites exhibit gradual loss in stiffness whereas the bidirectional woven composites show a relatively quicker loss during stage II of fatigue damage progression. Both, the static and the fatigue failures in unidirectional carbon fiber reinforced polymer composites originates due to generation of cracks on compression face while in bidirectional woven composites the damage ensues from both the compression and the tensile faces. These observations are supported by a detailed fractographic analysis.

Mycobacterium tuberculosis RecG Protein but Not RuvAB or RecA Protein Is Efficient at Remodeling the Stalled Replication Forks IMPLICATIONS FOR MULTIPLE MECHANISMS OF REPLICATION RESTART IN MYCOBACTERIA

Aberrant DNA replication, defects in the protection, and restart of stalled replication forks are major causes of genome instability in all organisms. Replication fork reversal is emerging as an evolutionarily conserved physiological response for restart of stalled forks. Escherichia coli RecG, RuvAB, and RecA proteins have been shown to reverse the model replication fork structures in vitro. However, the pathways and the mechanisms by which Mycobacterium tuberculosis, a slow growing human pathogen, responds to different types of replication stress and DNA damage are unclear. Here, we show that M. tuberculosis RecG rescues E. coli Delta recG cells from replicative stress. The purified M. tuberculosis RecG (MtRecG) and RuvAB(MtRuvAB) proteins catalyze fork reversal of model replication fork structures with and without a leading strand single-stranded DNA gap. Interestingly, single-stranded DNA-binding protein suppresses the MtRecG- and MtRuvAB-mediated fork reversal with substrates that contain lagging strand gap. Notably, our comparative studies with fork structures containing template damage and template switching mechanism of lesion bypass reveal that MtRecG but not MtRuvAB or MtRecA is proficient in driving the fork reversal. Finally, unlike MtRuvAB, we find that MtRecG drives efficient reversal of forks when fork structures are tightly bound by protein. These results provide direct evidence and valuable insights into the underlying mechanism of MtRecG-catalyzed replication fork remodeling and restart pathways in vivo.

AN INVESTIGATION OF CUTTING MECHANISMS AND STRAIN FIELDS DURING ORTHOGONAL CUTTING IN CFRPs

Fiber-reinforced plastics (FRPs) are typically difficult to machine due to their highly heterogeneous and anisotropic nature and the presence of two phases (fiber and matrix) with vastly different strengths and stiffnesses. Typical machining damage mechanisms in FRPs include series of brittle fractures (especially for thermosets) due to shearing and cracking of matrix material, fiber pull-outs, burring, fuzzing, fiber-matrix debonding, etc. With the aim of understanding the influence of the pronounced heterogeneity and anisotropy observed in FRPs, ``Idealized'' Carbon FRP (I-CFRP) plates were prepared using epoxy resin with embedded equispaced tows of carbon fibers. Orthogonal cutting of these I-CFRPs was carried out, and the chip formation characteristics, cutting force signals and strain distributions obtained during machining were analyzed using the Digital Image Correlation (DIC) technique. In addition, the same procedure was repeated on Uni-Directional CFRPs (UD-CFRPs). Chip formation mechanisms in FRPs were found to depend on the depth of cut and fiber orientation with pure epoxy showing a pronounced ``size effect.'' Experimental results indicate that in-situ full field strain measurements from DIC coupled with force measurements using dynamometry provide an adequate measure of anisotropy and heterogeneity during orthogonal cutting.