Static Optimizations Research Articles

A memory footprint optimization framework for Python applications targeting edge devices

The advantages of processing data at the source motivate developers to offload computations at the edges of IoT networks. However, the computational resource constraints of edge computing devices limit the opportunities for deploying applications developed in high-level languages at the edge. In contrast to C/C++, applications developed in Python and other dynamic high-level languages, often require an increased amount of memory size, due to their inherent static memory management approach. Therefore, developers targeting the deployment of applications in dynamic high-level languages at the edge need support by methodologies and tools, which will allow these applications to exploit the limited memory resources efficiently. This work presents a memory optimization framework for applications developed in the Python programming language targeting edge devices. Aiming to avoid the inherent pitfalls of static memory management that Python’s integrated memory manager imposes, the framework targets the reduction of the required memory footprint, by integrating a set of static and dynamic optimizations. The evaluation results, based on a set of representative real-life benchmark suites and applications, show 64% average memory footprint reduction, over the CPython’s minimum baseline of 24 MB. Additionally, we investigate the impact of memory size reduction on execution time and energy consumption. The results show 51% lower execution time and 47.3% reduction in the energy consumed.

Enhancing Conditional Stalling to Boost Performance of Stream-Processing Logic With RAW Dependencies

Ambiguous read-after-Write (RAW) dependencies are omnipresent in multiple streaming applications, establishing hard to optimize bottlenecks. Considering actual input data, these may rarely be true dependencies. However, the increasingly used High-Level Synthesis (HLS) compilers must assume the worst-case scenario, as they rely on static optimizations. Conditional stalling is a simple yet impactful technique, useful even when conflicts are common. At the cost of a small area penalty, it allows improving (in some cases, by several times) the mean throughput of these systems. In this brief, we describe a high-frequency HLS implementation of the technique and examine its behavior as a function of input and architecture characteristics, with the goal of understanding when to use it and how to optimize throughput.

Entropic Effects on the Aqueous Microsolvation of Protonated Glycine and Protonated β-Alanine. Hybrid Density Functional Theory Born-Oppenheimer Molecular Dynamics Studies.

Recent low-temperature infrared-based experimental studies provided information about the effects of aqueous microsolvation on the intramolecular hydrogen bond of protonated glycine and β-alanine [J. Phys. Chem. A 2019, 123, 3355]. Here we address the temperature-dependent entropic effects on the aqueous microsolvation patterns of these protonated amino acids using the AAH+(H2O)n (n = 1-8) cluster model at 50 K and room temperature with Born-Oppenheimer molecular dynamics using a calibrated hybrid density functional. The CCOOH-Ow, N-Ow, and center-of-mass-Ow radial distribution functions provide accurate structural data and temperature-dependent water coordination numbers vs. solvation degree. The solvation patterns for protonated glycine at 50 K show structural features in agreement with previous static optimizations. However, entropic effects at room temperature play a crucial role in the evolution of the intramolecular HB strength vs. solvation degree for both protonated amino acids. With increasing hydration entropic effects favor the making of solvent hydrogen bond networks over full solvation of protonated glycine. At room temperature four water molecules are needed to build the first solvation shell for protonated glycine while five are required for protonated β-alanine. A new statistical Cumulative Percentage of Structures (CPS) scheme is proposed; when the CPS data are analyzed in light of the empirical formula of Rozenberg et al. [Phys. Chem. Chem. Phys. 2000, 2, 2699] and the hydrogen bond relative strength (HBRS) criteria of Jeffrey [An Introduction to Hydrogen Bonding; Oxford University: 1997] we can provide a detailed molecular mechanism for the weakening of the intramolecular hydrogen bond based on the average dynamical structures, which clearly reveals the temperature dependence of this process. The new CPS-HBRS scheme proposed here can be utilized using any type of molecular dynamics trajectory (classical, BOMD, CPMD, etc.).

The road not taken: exploring alias analysis based optimizations missed by the compiler

Context-sensitive inter-procedural alias analyses are more precise than intra-procedural alias analyses. However, context-sensitive inter-procedural alias analyses are not scalable. As a consequence, most of the production compilers sacrifice precision for scalability and implement intra-procedural alias analysis. The alias analysis is used by many compiler optimizations, including loop transformations. Due to the imprecision of alias analysis, the program’s performance may suffer, especially in the presence of loops. Previous work proposed a general approach based on code-versioning with dynamic checks to disambiguate pointers at runtime. However, the overhead of dynamic checks in this approach is O(log n), which is substantially high to enable interesting optimizations. Other suggested approaches, e.g., polyhedral and symbolic range analysis, have O(1) overheads, but they only work for loops with certain constraints. The production compilers, such as LLVM and GCC, use scalar evolution analysis to compute an O(1) range check for loops to resolve memory dependencies at runtime. However, this approach also can only be applied to loops with certain constraints. In this work, we present our tool, Scout, that can disambiguate two pointers at runtime using single memory access. Scout is based on the key idea to constrain the allocation size and alignment during memory allocations. Scout can also disambiguate array accesses within a loop for which the existing O(1) range checks technique cannot be applied. In addition, Scout uses feedback from static optimizations to reduce the number of dynamic checks needed for optimizations. Our technique enabled new opportunities for loop-invariant code motion, dead store elimination, loop vectorization, and load elimination in an already optimized code. Our performance improvements are up to 51.11% for Polybench and up to 0.89% for CPU SPEC 2017 suites. The geometric means for our allocator’s CPU and memory overheads for CPU SPEC 2017 benchmarks are 1.05%, and 7.47%, respectively. For Polybench benchmarks, the geometric mean of CPU and memory overheads are 0.21% and 0.13%, respectively.

Cell-free biosynthesis meets dynamic optimization and control: a fed-batch framework

Cell-free biosynthesis uses the machinery of cells, such as the metabolic reactions, to carry out conversion processes in vitro. This can be more beneficial than in vivo approaches like fermentations. Some advantages of these synthetic biology processes include higher product yields, rates and titers, and more flexibility in pathway design. Cell-free biosynthesis is still in early stages and, unlike in vivo production, there are very few examples of model-based optimization. Moreover, we encounter static optimizations in most cases, neglecting the dynamic nature of the processes. We present an optimal control framework to maximize the efficiency of cell-free biosynthesis. We focus on fed-batch setups as they allow enhancing the reaction rates via the feeding, extending production processes for longer times, and minimizing the potential negative effects of enzyme kinetics with substrate inhibition. Our framework can in principle handle several cost functions and exploit both static and dynamic degrees of freedom. An aspect that can hinder model-based optimization is model uncertainty, which can arise due to uncertain parameters, oversimplified model assumptions or unknown reaction mechanisms. To counteract this, we propose the use of model predictive control during the process operation. In addition, we outline the use of moving horizon estimation as an observer in the case of unmeasured states. We consider the de novo cell-free synthesis of uridine diphosphate-N-acetylglucosamine as a biomedical relevant case study, where we were able to maximize the volumetric productivity in simulations, and indirectly also the titer and enzyme efficiency use.

Tf.data

Training machine learning models requires feeding input data for models to ingest. Input pipelines for machine learning jobs are often challenging to implement efficiently as they require reading large volumes of data, applying complex transformations, and transferring data to hardware accelerators while overlapping computation and communication to achieve optimal performance. We present tf.data, a framework for building and executing efficient input pipelines for machine learning jobs. The tf.data API provides operators that can be parameterized with user-defined computation, composed, and reused across different machine learning domains. These abstractions enable users to focus on the application logic of data processing, while tf.data's runtime ensures that pipelines run efficiently. We demonstrate that input pipeline performance is critical to the end-to-end training time of state-of-the-art machine learning models. tf.data delivers the high performance required, while avoiding the need for manual tuning of performance knobs. We show that tf.data features, such as parallelism, caching, static optimizations, and optional non-deterministic execution are essential for high performance. Finally, we characterize machine learning input pipelines for millions of jobs that ran in Google's datacenter fleet, showing that input data processing is highly diverse and consumes a significant fraction of job resources. Our analysis motivates future research directions, such as sharing computation across jobs and pushing data projection to the storage layer.

Sound garbage collection for C using pointer provenance

Garbage collection (GC) support for unmanaged languages can reduce programming burden in reasoning about liveness of dynamic objects. It also avoids temporal memory safety violations and memory leaks. Sound GC for weakly-typed languages such as C/C++, however, remains an unsolved problem. Current value-based GC solutions examine values of memory locations to discover the pointers, and the objects they point to. The approach is inherently unsound in the presence of arbitrary type casts and pointer manipulations, which are legal in C/C++. Such language features are regularly used, especially in low-level systems code. In this paper, we propose Dynamic Pointer Provenance Tracking to realize sound GC. We observe that pointers cannot be created out-of-thin-air, and they must have provenance to at least one valid allocation. Therefore, by tracking pointer provenance from the source (e.g., malloc) through both explicit data-flow and implicit control-flow, our GC has sound and precise information to compute the set of all reachable objects at any program state. We discuss several static analysis optimizations, that can be employed during compilation aided with profiling, to significantly reduce the overhead of dynamic provenance tracking from nearly 8× to 16% for well-behaved programs that adhere to the C standards. Pointer provenance based sound GC invocation is also 13% faster and reclaims 6% more memory on average, compared to an unsound value-based GC.

Design and Optimization of an Aeroservoelastic Wind Tunnel Model

Through the combination of passive and active load alleviation techniques, this paper presents the design, optimization, manufacturing, and update of a flexible composite wind tunnel model. In a first step, starting from the specification of an adequate wing and trailing edge flap geometry, passive, static aeroelastic stiffness optimizations for various objective functions have been performed. The second optimization step comprised a discretization of the continuous stiffness distributions, resulting in manufacturable stacking sequences. In order to determine which of the objective functions investigated in the passive structural optimization most efficiently complemented the projected active control schemes, the condensed modal finite element models were integrated in an aeroelastic model, involving a dedicated gust load alleviation controller. The most promising design was selected for manufacturing. The finite element representation could be updated to conform to the measured eigenfrequencies, based on the dynamic identification of the model. Eventually, a wind tunnel test campaign was conducted in November 2018 and results have been examined in separate reports.

Aqueous solvation of Mg(ii) and Ca(ii): A Born-Oppenheimer molecular dynamics study of microhydrated gas phase clusters.

The hydration features of [Mg(H2O)n]2+ and [Ca(H2O)n]2+ clusters with n = 3-6, 8, 18, and 27 were studied by means of Born-Oppenheimer molecular dynamics simulations at the B3LYP/6-31+G** level of theory. For both ions, it is energetically more favorable to have all water molecules in the first hydration shell when n ≤ 6, but stable lower coordination average structures with one water molecule not directly interacting with the ion were found for Mg2+ at room temperature, showing signatures of proton transfer events for the smaller cation but not for the larger one. A more rigid octahedral-type structure for Mg2+ than for Ca2+ was observed in all simulations, with no exchange of water molecules to the second hydration shell. Significant thermal effects on the average structure of clusters were found: while static optimizations lead to compact, spherically symmetric hydration geometries, the effects introduced by finite-temperature dynamics yield more prolate configurations. The calculated vibrational spectra are in agreement with infrared spectroscopy results. Previous studies proposed an increase in the coordination number (CN) from six to eight water molecules for [Ca(H2O)n]2+ clusters when n ≥ 12; however, in agreement with recent measurements of binding energies, no transition to a larger CN was found when n > 8. Moreover, the excellent agreement found between the calculated extended X-ray absorption fine structure spectroscopy spectra for the larger cluster and the experimental data of the aqueous solution supports a CN of six for Ca2+.

A computational framework for simultaneous estimation of muscle and joint contact forces and body motion using optimization and surrogate modeling

Concurrent estimation of muscle activations, joint contact forces, and joint kinematics by means of gradient-based optimization of musculoskeletal models is hindered by computationally expensive and non-smooth joint contact and muscle wrapping algorithms. We present a framework that simultaneously speeds up computation and removes sources of non-smoothness from muscle force optimizations using a combination of parallelization and surrogate modeling, with special emphasis on a novel method for modeling joint contact as a surrogate model of a static analysis. The approach allows one to efficiently introduce elastic joint contact models within static and dynamic optimizations of human motion. We demonstrate the approach by performing two optimizations, one static and one dynamic, using a pelvis-leg musculoskeletal model undergoing a gait cycle. We observed convergence on the order of seconds for a static optimization time frame and on the order of minutes for an entire dynamic optimization. The presented framework may facilitate model-based efforts to predict how planned surgical or rehabilitation interventions will affect post-treatment joint and muscle function.

The Dinamica EGO virtual machine

This paper describes DinamicaVM, the virtual machine that runs applications developed in Dinamica EGO. Dinamica EGO is a framework used in the development of geomodeling applications. Behind its multitude of visual modes and graphic elements, Dinamica EGO runs on top of a virtual machine. This machine – DinamicaVM – offers developers a high-level instruction set architecture, featuring elements such as map and reduce, which are typical in the functional/parallel world. Ensuring that these very expressive components work together efficiently is a challenging endeavor. Dinamica's runtime addresses this challenge through a suite of optimizations, which borrows ideas from functional programming languages, and leverages specific behavior expected in geo-scientific programs. As we show in this paper some of these optimizations deliver speedups of almost 100×, and are key to the industrial-quality performance of one of the world's most widely used geomodeling tools. • Description of virtual machine widely known among geoscientists. • High-level instruction set that uses functional components as building blocks. • Several static analyses and optimizations that boost the VM's performance. • Detailed description of the copy elimination problem, which we solve.

Online Aggregation of the Forwarding Information Base: Accounting for Locality and Churn

This paper studies the problem of compressing the forwarding information base (FIB), but taking a wider perspective. Indeed, FIB compression goes beyond sheer compression, as the gain in memory use obtained from the compression has consequences on the updates that will have to be applied to the compressed FIB. We are interested in the situation where forwarding rules can change over time, e.g., due to border gateway protocol (BGP) route updates. Accordingly, we frame FIB compression as an online problem and design competitive online algorithms to solve it. In contrast to prior work which mostly focused on static optimizations, we study an online variant of the problem where routes can change over time and where the number of updates to the FIB is taken into account explicitly. The reason to consider this version of the problem is that leveraging temporal locality while accounting for the number of FIB updates helps to keep routers CPU load low and reduces the number of FIB updates to be transferred, e.g., from the network-attached software-defined network controller to a remote switch. This paper introduces a formal model which is an interesting generalization of several classic online aggregation problems. Our main contribution is an O(w) -competitive algorithm, where ${w}$ is the length of an IP address. We also derive a lower bound which shows that our result is asymptotically optimal within a natural class of algorithms, based on so-called sticks .

Surface irrigation simulation-optimization model based on meta-heuristic algorithms

Simulation-optimization models are valuable tools for determining the optimal performance of systems. The main objective of this research was to develop and validate EDOSIM (Evaluation, Design, and Optimization of Surface Irrigation Model) as a simulation-optimization model for surface irrigation systems. For simulation, which consisted of the design or evaluation of basin, border and furrow irrigation, the Volume Balance model was used. For optimization, twenty meta-heuristic algorithms were applied. In this model, based on irrigation, the volume of infiltrated water to soil was calculated without having advance and recession data. The hydraulic objective function was used to minimize the linear combination of seven performance indicators. Regarding the optimization of the objective function, the functional, multi-dimensional, static, constraint, continuous, single-objective, and meta-heuristic optimizations were applied. Data obtained from fifteen experimental fields were used for the validation of simulation, algorithms parameters setting, and validation of optimization. Comparison of the simulation results of the EDOSIM model with those of the Hydrodynamic model of SIRMOD software showed the good performance of EDOSIM model and the proposed method for estimating the volume of infiltration with RMSE = 0.068, R2 = 0.988, CRM = 0.005 and NRMSE = 4.2%. The Shuffled Complex Evolution (SCE) algorithm was found to be the best algorithm for the optimization of fields; in all fields, the objective function was decreased (improved). Comparison of the objective function of the EDOSIM model with eight solvers of Optimization and Global Optimization Toolboxes of MATLAB software also revealed the superiority of the EDOSIM model for optimization.

Muscle function in glenohumeral joint stability during lifting task.

Ensuring glenohumeral stability during repetitive lifting tasks is a key factor to reduce the risk of shoulder injuries. Nevertheless, the literature reveals some lack concerning the assessment of the muscles that ensure glenohumeral stability during specific lifting tasks. Therefore, the purpose of this study was to assess the stabilization function of shoulder muscles during a lifting task. Kinematics and muscle electromyograms (n = 9) were recorded from 13 healthy adults during a bi-manual lifting task performed from the hip to the shoulder level. A generic upper-limb OpenSim model was implemented to simulate glenohumeral stability and instability by performing static optimizations with and without glenohumeral stability constraints. This procedure enabled to compute the level of shoulder muscle activity and forces in the two conditions. Without the stability constraint, the simulated movement was unstable during 74%±16% of the time. The force of the supraspinatus was significantly increased of 107% (p<0.002) when the glenohumeral stability constraint was implemented. The increased supraspinatus force led to greater compressive force (p<0.001) and smaller shear force (p<0.001), which contributed to improved glenohumeral stability. It was concluded that the supraspinatus may be the main contributor to glenohumeral stability during lifting task.

Ab initio study of the dolomite to dolomite-II high-pressure phase transition

Ab initio calculations were performed on ordered and disordered dolomite and dolomite-II crystal structures. The mechanism for the dolomite to dolomite-II phase transition was investigated by the calculation of vibrational frequencies. In particular, a soft mode was observed at the F point of the rhombohedral Brillouin zone, whose frequency becomes imaginary at pressure ( P ) higher than 16.70 GPa. Static geometry optimizations of dolomite-II in the P range 0–32 GPa allowed to study both the dolomite structural evolution with P and its compressibility. As pressure decreases, the dolomite-II distortion decreases as well, becoming almost regular at ∼17 GPa (phase transition) and increasing again after the phase transition, to values comparable to the low-pressure dolomite. The deformation style slightly changes in the dolomite-II like crystal structures before and after the phase transition. At low P (dolomite), K = 95.4 (5) GPa and K ′ = 4.26(8), whereas dolomite-II after the phase transition has K ′ = 3.44(3), by fixing V and K at the value obtained for dolomite at the equilibrium. The influence of cation disorder on the baric behavior of dolomite was also investigated. No phase transition was predicted for disordered dolomite, at least up to 26 GPa. Thermodynamic calculations carried out on the two dolomite polymorphs allow the evaluation of temperature ( T ) effects on the phase transition in the range of validity of a fully ordered structural model. The dolomite to dolomite-II phase transition boundary is located at P = 16.75 GPa at T = 300 K, in agreement with considerations based on the static calculation and the analysis of the soft mode.

Dynamic Power and Energy Management for Energy Harvesting Nonvolatile Processor Systems

Self-powered systems running on scavenged energy will be a key enabler for pervasive computing across the Internet of Things. The variability of input power in energy-harvesting systems limits the effectiveness of static optimizations aimed at maximizing the input-energy-to-computation ratio. We show that the resultant gap between available and exploitable energy is significant, and that energy storage optimizations alone do not significantly close the gap. We characterize these effects on a real, fabricated energy-harvesting system based on a nonvolatile processor. We introduce a unified energy-oriented approach to first optimize the number of backups, by more aggressively using the stored energy available when power failure occurs, and then optimize forward progress via improving the rate of input energy to computation via dynamic voltage and frequency scaling and self-learning techniques. We evaluate combining these schemes and show capture of up to 75.5% of all input energy toward processor computation, an average of 1.54 × increase over the best static “Forward Progress” baseline system. Notably, our energy-optimizing policy combinations simultaneously improve both the rate of forward progress and the rate of backup events (by up to 60.7% and 79.2% for RF power, respectively, and up to 231.2% and reduced to zero, respectively, for solar power). This contrasts with static frequency optimization approaches in which these two metrics are antagonistic.

Effect of lateralized design on muscle and joint reaction forces for reverse shoulder arthroplasty

Manufacturers of reverse shoulder arthroplasty (RSA) implants have recently designed innovative implants to optimize performance in rotator cuff-deficient shoulders. These advancements are not without tradeoffs and can have negative biomechanical effects. The objective of this study was to develop an integrated finite element analysis-kinematic model to compare the muscle forces and joint reaction forces (JRFs) of 3 different RSA designs. A kinematic model of a normal shoulder joint was adapted from the Delft model and integrated with the well-validated OpenSim shoulder model. Static optimizations then allowed for calculation of the individual muscle forces, moment arms, and JRFs relative to net joint moments. Three-dimensional computer models of 3 RSA designs-humeral lateralized design (HLD), glenoid lateralized design, and Grammont design-were integrated, and parametric studies were performed. Overall, there were decreases in deltoid and rotator cuff muscle forces for all 3 RSA designs. These decreases were greatest in the middle deltoid of the HLD model for abduction and flexion and in the rotator cuff muscles under both internal rotation and external rotation. The JRFs in abduction and flexion decreased similarly for all RSA designs compared with the normal shoulder model, with the greatest decrease seen in the HLD model. These findings demonstrate that the design characteristics implicit in these modified RSA prostheses result in mechanical differences most prominently seen in the deltoid muscle and overall JRFs. Further research using this novel integrated model can help guide continued optimization of RSA design and clinical outcomes.

Musculoskeletal Modeling of the Lumbar Spine to Explore Functional Interactions between Back Muscle Loads and Intervertebral Disk Multiphysics.

During daily activities, complex biomechanical interactions influence the biophysical regulation of intervertebral disks (IVDs), and transfers of mechanical loads are largely controlled by the stabilizing action of spine muscles. Muscle and other internal forces cannot be easily measured directly in the lumbar spine. Hence, biomechanical models are important tools for the evaluation of the loads in those tissues involved in low-back disorders. Muscle force estimations in most musculoskeletal models mainly rely, however, on inverse calculations and static optimizations that limit the predictive power of the numerical calculations. In order to contribute to the development of predictive systems, we coupled a predictive muscle model with the passive resistance of the spine tissues, in a L3–S1 musculoskeletal finite element model with osmo-poromechanical IVD descriptions. The model included 46 fascicles of the major back muscles that act on the lower spine. The muscle model interacted with activity-related loads imposed to the osteoligamentous structure, as standing position and night rest were simulated through distributed upper body mass and free IVD swelling, respectively. Calculations led to intradiscal pressure values within ranges of values measured in vivo. Disk swelling led to muscle activation and muscle force distributions that seemed particularly appropriate to counterbalance the anterior body mass effect in standing. Our simulations pointed out a likely existence of a functional balance between stretch-induced muscle activation and IVD multiphysics toward improved mechanical stability of the lumbar spine understanding. This balance suggests that proper night rest contributes to mechanically strengthen the spine during day activity.

Development of an open access tool for design, simulated dispatch, and economic assessment of distributed generation technologies

The design and deployment of DG systems requires an integrated assessment of the building and generator dynamics including the time-variant energy costs and emission factors. Static design optimizations are unable to consider the physical generator operating constraints, seasonal variability and non-coincidence in electric, heating, and cooling demands. This paper introduces the Distributed Generation Build-out Economic Assessment Tool (DG-BEAT) which combines building, utilities, and emissions databases with a library of simplified generator and building models in a user-friendly interface. Five control strategies are presented for the dynamic dispatch of distributed generation technologies at commercial buildings. The control approaches stem from the physical limitations of different generator types. Methods are also outlined for the dispatch of complementary technologies (e.g. energy storage) and accommodation of on-site renewables (e.g. solar PV) which could further improve the economic or environmental benefits of distributed generation. This paper details the methodology of sizing and dispatching distributed generation components, outlines eight databases that are employed to capture regional variations in pricing and building dynamics, and discusses the myriad of customizations available to provide a tailored analysis for a single building or national impact studies.

Topology optimization for nonlinear structural problems based on artificial bee colony algorithm

Topology algorithm is suggested for nonlinear structural problems based on artificial bee colony algorithm (ABCA). Although ABCA has been successfully applied to static and dynamic stiffness topology optimizations, it has never been applied to nonlinear structural problems. In order to examine whether the topology algorithm is suitable for nonlinear problems, it is applied to geometrically nonlinear, materially nonlinear, and both geometrically nonlinear and materially nonlinear topology optimization. Waggle index update rule and changing filter scheme were implemented with ABCA to obtain a robust and stable optimized topology. Some examples were presented to show the applicability and effectiveness of the suggested algorithm and compared with solid isotropic material with penalization (SIMP). It was concluded that the proposed algorithm can be applied to the above three kinds of nonlinear structural problems and has been verified as effective and applicable topology algorithm.