Large Data Structures Research Articles

Resurrector

Modern object-oriented applications commonly suffer from severe performance problems that need to be optimized away for increased efficiency and user satisfaction. Many existing optimization techniques (such as object pooling and pretenuring) require precise identification of object lifetimes. However, it is particularly challenging to obtain object lifetimes both precisely and efficiently: precise profiling techniques such as Merlin introduce several hundred times slowdown even for small programs while efficient approximation techniques often sacrifice precision and produce less useful lifetime information. This paper presents a tunable profiling technique, called Resurrector , that explores the middle ground between high precision and high efficiency to find the precision-efficiency sweetspot for various livenessbased optimization techniques. Our evaluation shows that Resurrector is both more precise and more efficient than the GC-based approximation, and it is orders-of-magnitude faster than Merlin. To demonstrate Resurrector's usefulness, we have developed client analyses to find allocation sites that create large data structures with disjoint lifetimes. By inspecting program source code and reusing data structures created from these allocation sites, we have achieved significant performance gains. We have also improved the precision of an existing optimization technique using the lifetime information collected by Resurrector.

A novel architecture for ahead branch prediction

In theory, branch predictors with more complicated algorithms and larger data structures provide more accurate predictions. Unfortunately, overly large structures and excessively complicated algorithms cannot be implemented because of their long access delay. To date, many strategies have been proposed to balance delay with accuracy, but none has completely solved the issue. The architecture for ahead branch prediction (A2BP) separates traditional predictors into two parts. First is a small table located at the front-end of the pipeline, which makes the prediction brief enough even for some aggressive processors. Second, operations on complicated algorithms and large data structures for accurate predictions are all moved to the back-end of the pipeline. An effective mechanism is introduced for ahead branch prediction in the back-end and small table update in the front. To substantially improve prediction accuracy, an indirect branch prediction algorithm based on branch history and target path (BHTP) is implemented in A2BP. Experiments with the standard performance evaluation corporation (SPEC) benchmarks on gem5/SimpleScalar simulators demonstrate that A2BP improves average performance by 2.92% compared with a commonly used branch target buffer-based predictor. In addition, indirect branch misses with the BHTP algorithm are reduced by an average of 28.98% compared with the traditional algorithm.

Fault Tolerance in Distributed Systems Using Fused Data Structures

Replication is the prevalent solution to tolerate faults in large data structures hosted on distributed servers. To tolerate f crash faults (dead/unresponsive data structures) among n distinct data structures, replication requires f + 1 replicas of each data structure, resulting in nf additional backups. We present a solution, referred to as fusion that uses a combination of erasure codes and selective replication to tolerate f crash faults using just f additional fused backups. We show that our solution achieves O(n) savings in space over replication. Further, we present a solution to tolerate f Byzantine faults (malicious data structures), that requires only nf + f backups as compared to the 2nf backups required by replication. We explore the theory of fused backups and provide a library of such backups for all the data structures in the Java Collection Framework. The theoretical and experimental evaluation confirms that the fused backups are space-efficient as compared to replication, while they cause very little overhead for normal operation. To illustrate the practical usefulness of fusion, we use fused backups for reliability in Amazon's highly available key-value store, Dynamo. While the current replication-based solution uses 300 backup structures, we present a solution that only requires 120 backup structures. This results in savings in space as well as other resources such as power.

DSK: k-mer counting with very low memory usage

Counting all the k-mers (substrings of length k) in DNA/RNA sequencing reads is the preliminary step of many bioinformatics applications. However, state of the art k-mer counting methods require that a large data structure resides in memory. Such structure typically grows with the number of distinct k-mers to count. We present a new streaming algorithm for k-mer counting, called DSK (disk streaming of k-mers), which only requires a fixed user-defined amount of memory and disk space. This approach realizes a memory, time and disk trade-off. The multi-set of all k-mers present in the reads is partitioned, and partitions are saved to disk. Then, each partition is separately loaded in memory in a temporary hash table. The k-mer counts are returned by traversing each hash table. Low-abundance k-mers are optionally filtered. DSK is the first approach that is able to count all the 27-mers of a human genome dataset using only 4.0 GB of memory and moderate disk space (160 GB), in 17.9 h. DSK can replace a popular k-mer counting software (Jellyfish) on small-memory servers. http://minia.genouest.org/dsk

Chaperones and impersonators

Chaperones and impersonators provide run-time support for interposing on primitive operations such as function calls, array access and update, and structure field access and update. Unlike most interposition support, chaperones and impersonators are restricted so that they constrain the behavior of the interposing code to reasonable interposition, which in practice preserves the abstraction mechanisms and reasoning that programmers and compiler analyses rely on. Chaperones and impersonators are particularly useful for implementing contracts, and our implementation in Racket allows us to improve both the expressiveness and the performance of Racket's contract system. Specifically, contracts on mutable data can be enforced without changing the API to that data; contracts on large data structures can be checked lazily on only the accessed parts of the structure; contracts on objects and classes can be implemented with lower overhead; and contract wrappers can preserve object equality where appropriate. With this extension, gradual typing systems, such as Typed Racket, that rely on contracts for interoperation with untyped code can now pass mutable values safely between typed and untyped modules.

Robust Pipelined Memory System with Worst Case Performance Guarantee for Network Processing

Many network processing applications require wirespeed access to large data structures or a large amount of packet and flow-level data. Therefore, it is essential for the memory system of a router to be able to support both read and write accesses to such data at link speeds. As link speeds continue to increase, router designers are constantly grappling with the unfortunate trade-offs between the speed and cost of SRAM and DRAM. The capacity of SRAMs is woefully inadequate in many cases and it proves too costly to store large data structures entirely in SRAM, while DRAM is viewed as too slow for providing wirespeed updates at such high speed. In this paper, we analyze a robust pipelined memory architecture that can emulate an ideal SRAM by guaranteeing with very high probability that the output sequence produced by the pipelined memory architecture is the same as the one produced by an ideal SRAM under the same sequence of memory read and write operations, except time shifted by a fixed pipeline delay of \Delta. Given a fixed pipeline delay abstraction, no interrupt mechanism is required to indicate when read data are ready or a write operation has completed, which greatly simplifies the use of the proposed solution. The design is based on the interleaving of DRAM banks together with the use of a reservation table that serves in part as a data cache. In contrast to prior interleaved memory solutions, our design is robust under all memory access patterns, including adversarial ones, which we demonstrate through a rigorous worst case theoretical analysis using a combination of convex ordering and large deviation theory.

FlashTrie: Beyond 100-Gb/s IP Route Lookup Using Hash-Based Prefix-Compressed Trie

It is becoming apparent that the next-generation IP route lookup architecture needs to achieve speeds of 100 Gb/s and beyond while supporting IPv4 and IPv6 with fast real-time updates to accommodate ever-growing routing tables. Some of the proposed multibit-trie-based schemes, such as TreeBitmap, have been used in today's high-end routers. However, their large data structures often require multiple external memory accesses for each route lookup. A pipelining technique is widely used to achieve high-speed lookup with the cost of using many external memory chips. Pipelining also often leads to poor memory load-balancing. In this paper, we propose a new IP route lookup architecture called FlashTrie that overcomes the shortcomings of the multibit-trie-based approaches. We use a hash-based membership query to limit off-chip memory accesses per lookup and to balance memory utilization among the memory modules. By compacting the data structure size, the lookup depth of each level can be increased. We also develop a new data structure called Prefix-Compressed Trie that reduces the size of a bitmap by more than 80%. Our simulation and implementation results show that FlashTrie can achieve 80-Gb/s worst-case throughput while simultaneously supporting 2 M prefixes for IPv4 and 318 k prefixes for IPv6 with one lookup engine and two Double-Data-Rate (DDR3) SDRAM chips. When implementing five lookup engines on a state-of-the-art field programmable gate array (FPGA) chip and using 10 DDR3 memory chips, we expect FlashTrie to achieve 1-Gpps (packet per second) throughput, equivalent to 400 Gb/s for IPv4 and 600 Gb/s for IPv6. FlashTrie also supports incremental real-time updates.

Application of Time Series Hyperspectral Imaging (TS-HSI) for Determining Water Distribution Within Beef and Spectral Kinetic Analysis During Dehydration

This study was carried out for rapid and noninvasive determination of water distribution within beef during dehydration using time series hyperspectral imaging (TS-HSI). Hyperspectral images (380–1,700 nm) of beef slices were acquired at different periods of dehydration process. The spectra of beef were extracted from the TS-HSI images using image segmentation process. Principal component analysis was conducted to obtain an overview of the systematic spectral variations during dehydration. Instead of the traditional data mining strategies to cope with the large multivariate data structures in the TS-HSI images, the selection of effective wavelengths was conducted for the first time to reduce the computational burden of the TS-HSI data and predigest calibration modeling. On the basis of the effective wavelengths identified by using successive projections algorithm (SPA), three spectral calibration algorithms of partial least squares regression, least squares support vector machines, and multiple linear regression (MLR) were compared. The SPA-MLR model with Spectral Set I was considered to be the best for determining water content of beef slice. The model led to a coefficient of determination ( $$ r_V^2 $$ ) of 0.953 and root mean square error estimated by cross-validation of 1.280 %. The visualization of water distribution within beef slice during dehydration was finally generated by transferring the quantitative model to each pixel in the image to determine water content in all spots of the beef sample. Kinetic analysis of the TS-HSI images was also conducted for the first time to analyze spectral changes of beef during dehydration. The results demonstrate that TS-HSI has the potential of quantitatively visualizing water content of beef rapidly and noninvasively during dehydration in a reasonable accuracy.

Bayesian analysis of zero inflated spatiotemporal HIV/TB child mortality data through the INLA and SPDE approaches: Applied to data observed between 1992 and 2010 in rural North East South Africa

Longitudinal mortality data with few deaths usually have problems of zero-inflation. This paper presents and applies two Bayesian models which cater for zero-inflation, spatial and temporal random effects. To reduce the computational burden experienced when a large number of geo-locations are treated as a Gaussian field (GF) we transformed the field to a Gaussian Markov Random Fields (GMRF) by triangulation. We then modelled the spatial random effects using the Stochastic Partial Differential Equations (SPDEs). Inference was done using a computationally efficient alternative to Markov chain Monte Carlo (MCMC) called Integrated Nested Laplace Approximation (INLA) suited for GMRF. The models were applied to data from 71,057 children aged 0 to under 10 years from rural north-east South Africa living in 15,703 households over the years 1992–2010. We found protective effects on HIV/TB mortality due to greater birth weight, older age and more antenatal clinic visits during pregnancy (adjusted RR (95% CI)): 0.73(0.53;0.99), 0.18(0.14;0.22) and 0.96(0.94;0.97) respectively. Therefore childhood HIV/TB mortality could be reduced if mothers are better catered for during pregnancy as this can reduce mother-to-child transmissions and contribute to improved birth weights. The INLA and SPDE approaches are computationally good alternatives in modelling large multilevel spatiotemporal GMRF data structures.

Bridging the Gap [From the Guest Editors

This special issue of IEEE Signal Processing Magazine is an attempt to renew the dialogue between these two communities. This issue contains articles that highlight the application of signal processing methods to geophysical problems. Geophysics is an entire interdisciplinary science that nurtures itself from the field of geology, mathematics, and physics. For years, special emphasis has been paid to geophysical data acquisition and data processing, as these two key aspects are an integral part of our efforts to estimate earth models and to monitor processes that occur deep within Earth's interior. At the core of acquisition and processing techniques is the problem of estimating models in the presence of noise and methods that permit us to assimilate and process large multidimensional data structures that depend on space and time.

Scalable parallel debugging with statistical assertions

Traditional debuggers are of limited value for modern scientific codes that manipulate large complex data structures. This paper discusses a novel debug-time assertion, called a "Statistical Assertion", that allows a user to reason about large data structures, and the primitives are parallelised to provide an efficient solution. We present the design and implementation of statistical assertions, and illustrate the debugging technique with a molecular dynamics simulation. We evaluate the performance of the tool on a 12,000 cores Cray XE6.

Debugging Scientific Applications With Statistical Assertions

Traditional debuggers are of limited value for modern scientific codes that manipulate large complex data structures. Current parallel machines make this even more complicated, because the data may be distributed across multiple processors, making it difficult to view, interpret and validate the contents of a distributed structure. As a result, many applications’ developers resort to placing validation and display code directly in the source program itself. This paper discusses a novel debug-time assertion, called a “Statistical Assertion”, that allows a user to reason about large data structures. We present the design and implementation of statistical assertions, and illustrate the debugging technique with a molecular dynamics simulation. We evaluate the performance of the system on a 12,000 processor Cray XE6, and show that it is useful for real time debugging.

Time series hyperspectral chemical imaging data: Challenges, solutions and applications

Hyperspectral chemical imaging (HCI) integrates imaging and spectroscopy resulting in three-dimensional data structures, hypercubes, with two spatial and one wavelength dimension. Each spatial image pixel in a hypercube contains a spectrum with >100 datapoints. While HCI facilitates enhanced monitoring of multi-component systems; time series HCI offers the possibility of a more comprehensive understanding of the dynamics of such systems and processes. This implies a need for modeling strategies that can cope with the large multivariate data structures generated in time series HCI experiments. The challenges posed by such data include dimensionality reduction, temporal morphological variation of samples and instrumental drift. This article presents potential solutions to these challenges, including multiway analysis, object tracking, multivariate curve resolution and non-linear regression. Several real world examples of time series HCI data are presented to illustrate the proposed solutions.

CUDA Memory Optimizations for Large Data-Structures in the Gravit Simulator

Modern GPUs open a completely new field to optimize embarrassingly parallel algorithms. Implementing an algorithm on a GPU confronts the programmer with a new set of challenges for program optimization. Especially tuning the program for the GPU memory hierarchy whose organization and performance implications are radically different from those of general purpose CPUs; and optimizing programs at the instruction-level for the GPU. In this paper we analyze different approaches for optimizing the memory usage and access patterns for GPUs and propose a class of memory layout optimizations that can take full advantage of the unique memory hierarchy of NVIDIA CUDA. Furthermore, we analyze some classical optimization techniques and how they effect the performance on a GPU. We used the Gravit gravity simulator to demonstrate these optimizations. The final optimized GPU version achieves a 87× speedup compared to the original CPU version. Almost 30% of this speedup are direct results of the optimizations discussed in this paper.

Improved version of parallel programming interface for distributed data with multiple helper servers

We present an improved version of the Parallel Programming Interface for Distributed Data with Multiple Helper Servers (PPIDDv2) library, which provides a common application programming interface that is based on the most frequently used functionality of both MPI-2 and GA. Compared with the previous version, the PPIDDv2 library introduces multiple helper servers to facilitate global data structures, and allows programmers to make heavy use of large global data structures efficiently. Program summaryProgram title: PPIDDv2Catalogue identifier: AEEF_v2_0Program summary URL:http://cpc.cs.qub.ac.uk/summaries/AEEF_v2_0.htmlProgram obtainable from: CPC Program Library, Queenʼs University, Belfast, N. IrelandLicensing provisions: Standard CPC licence, http://cpc.cs.qub.ac.uk/licence/licence.htmlNo. of lines in distributed program, including test data, etc.: 22 997No. of bytes in distributed program, including test data, etc.: 184 477Distribution format: tar.gzProgramming language: Fortran, CComputer: Many parallel systemsOperating system: VariousHas the code been vectorised or parallelised?: Yes. 2–1024 processors usedRAM: 50 MbytesClassification: 6.5External routines: Global Arrays or MPI-2Catalogue identifier of previous version: AEEF_v1_0Journal reference of previous version: Comput. Phys. Comm. 180 (2009) 2673Does the new version supersede the previous version?: YesNature of problem: Many scientific applications require management and communication of data that is global, and the standard MPI-2 protocol provides only low-level methods for the required one-sided remote memory access.Solution method: The Parallel Programming Interface for Distributed Data (PPIDD) library provides an interface, suitable for use in parallel scientific applications, that delivers communications and global data management. The library can be built either using the Global Arrays (GA) toolkit, or a standard MPI-2 library. This abstraction allows the programmer to write portable parallel codes that can utilise the best, or only, communications library that is available on a particular computing platform.Reasons for new version: In the previous version, functionality in global data structure was mainly implemented by MPI-2 passive one-sided operations. In real applications which make heavy use of global data structures, very poor performance was observed.Summary of revisions: Multiple helper servers are introduced to facilitate the manipulation and management of global data structure. Mutual exclusion is also implemented by the help of a data server, and becomes much more robust and efficient. In addition, flexible options are provided to choose different settings for helper servers. Significant improvement has been seen in performance tests.Running time: Problem-dependent. The test provided with the distribution takes only a few seconds to run.

Bayesian real-time perception algorithms on GPU

In this text we present the real-time implementation of a Bayesian framework for robotic multisensory perception on a graphics processing unit (GPU) using the Compute Unified Device Architecture (CUDA). As an additional objective, we intend to show the benefits of parallel computing for similar problems (i.e. probabilistic grid-based frameworks), and the user-friendly nature of CUDA as a programming tool. Inspired by the study of biological systems, several Bayesian inference algorithms for artificial perception have been proposed. Their high computational cost has been a prohibitory factor for real-time implementations. However in some cases the bottleneck is in the large data structures involved, rather than the Bayesian inference per se. We will demonstrate that the SIMD (single-instruction, multiple-data) features of GPUs provide a means for taking a complicated framework of relatively simple and highly parallelisable algorithms operating on large data structures, which might take up to several minutes of execution with a regular CPU implementation, and arrive at an implementation that executes in the order of tenths of a second. The implemented multimodal perception module (including stereovision, binaural sensing and inertial sensing) builds an egocentric representation of occupancy and local motion, the Bayesian Volumetric Map (BVM), based on which gaze shift decisions are made to perform active exploration and reduce the entropy of the BVM. Experimental results show that the real-time implementation successfully drives the robotic system to explore areas of the environment mapped with high uncertainty.

Using desktop computers to solve large-scale dense linear algebra problems

We provide experimental evidence that current desktop computers feature enough computational power to solve large-scale dense linear algebra problems. While the high computational cost of the numerical methods for solving these problems can be tackled by the multiple cores of current processors, we propose to use the disk to store the large data structures associated with these applications. Our results also show that the limited amount of RAM and the comparatively slow disk of the system pose no problem for the solution of very large dense linear systems and linear least-squares problems. Thus, current desktop computers are revealed as an appealing, cost-effective platform for research groups that have to deal with large dense linear algebra problems but have no direct access to large computing facilities.

Using SPARQL to Test for Lattices: application to quality assurance in biomedical ontologies.

We present a scalable, SPARQL-based computational pipeline for testing the lattice-theoretic properties of partial orders represented as RDF triples. The use case for this work is quality assurance in biomedical ontologies, one desirable property of which is conformance to lattice structures. At the core of our pipeline is the algorithm called NuMi, for detecting the Number of Minimal upper bounds of any pair of elements in a given finite partial order. Our technical contribution is the coding of NuMi completely in SPARQL. To show its scalability, we applied NuMi to the entirety of SNOMED CT, the largest clinical ontology (over 300,000 conepts). Our experimental results have been groundbreaking: for the first time, all non-lattice pairs in SNOMED CT have been identified exhaustively from 34 million candidate pairs using over 2.5 billion queries issued to Virtuoso. The percentage of non-lattice pairs ranges from 0 to 1.66 among the 19 SNOMED CT hierarchies. These non-lattice pairs represent target areas for focused curation by domain experts. RDF, SPARQL and related tooling provide an e cient platform for implementing lattice algorithms on large data structures.

Ternary CAM Power and Delay Model: Extensions and Uses

Applications in computer networks often require high throughput access to large data structures for lookup and classification. While advanced algorithms exist to speed these search primitives on network processors and even custom application-specific integrated circuits (ASICs), achieving tight bounds on worst case performance with standard memories often requires a very careful analysis of all possible access patterns. An alternative, and often times more simple solution, is possible if a ternary CAM (TCAM) is used to perform a fully parallel search across the entire data set. Unfortunately, this parallelism means that large portions of the chip are switching during each cycle, causing large amounts of power to be consumed. While researchers at all levels of design (from algorithms to circuits) have begun to explore new ways of managing the power consumption, quantifying design alternatives is difficult due to a lack of available models. In this paper, we examine the structure of a modern TCAM and present a simple, yet accurate, power and delay model. We present techniques to estimate the dynamic power consumption and leakage power of a TCAM structure and validate the model using a combination of industrial TCAM datasheets and prior published works. Such a model is a critical first step in bridging the intellectual divide between circuit-level and algorithm-level optimizations. To demonstrate the utility of our model, we present an extensive analysis of the model by varying various architectural parameters and describe how our model can be easily extended to handle several circuit optimizations in the TCAM structure. In addition, we present a comparative study of SRAM and TCAM energy consumption to directly quantify the many design options which will be very useful for network designers to explore various power management schemes.

Starc

Data structure corruptions are insidious bugs that reduce the reliability of software systems. Constraint-based datastructure repair promises to help programs recover from potentially crippling corruption errors. Prior work repairs a variety of relatively small data structures, usually with hundreds of nodes. We present STARC which uses static analysis to repair data structures with tens of thousands of nodes. Given a Java predicate method that describes the integrity constraints of a structure, STARC statically analyzes the method to identify: (1) the recurrent fields, i.e., fields that the predicate method uses to traverse the structure; and (2) local field constraints, i.e., how the value of an object field is related to the value of a neighboring object field. STARC executes the predicate method on the structure and monitors its execution to identify corrupt object fields, which STARC then repairs using a systematic search of a neighborhood of the given structure. Each repair action is guided by the result of the static analysis, which enables more efficient and effective repair compared to prior work. Experimental results showthat STARC can repair structures with tens of thousands of nodes, up to 100 times larger than prior work. STARC efficiency is probably not practical for very large data structures in deployed systems, but opens a promising direction for future work.