Search NASA⌕ Search

SEARCH · Search NASA

Results for “read performance”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 163 records · Page 9
All sourcesNASA NTRSDOE OSTI
DOE OSTI2017

Reproducibility and Variability of I/O Performance on BG/Q: Lessons Learned from a Data Aggregation Algorithm

Reading and writing data efficiently from different tiers of storage is necessary for most scientific simulations to achieve good performance at scale. Many software solutions have been developed to decrease the I/O bottleneck. One wellknown strategy, in the context of collective I/O operations, is the two-phase I/O scheme. This strategy consists of selecting a subset of processes to aggregate contiguous pieces of data before performing reads/writes. In our previous work, we implemented the two-phase I/O scheme with a MPI-based topology-aware algorithm. Our algorithm showed very good performance at scale compared to the standard I/O libraries such as POSIX I/O and MPI I/O. However, the algorithm had several limitations hindering a satisfying reproducibility of our experiments. In this paper, we extend our work by 1) identifying the obstacles we face to reproduce our experiments and 2) discovering solutions that reduce the unpredictability of our results.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2019

A Zoom-in Analysis of I/O Logs to Detect Root Causes of I/O Performance Bottlenecks

Scientific applications frequently spend a large fraction of their execution time in reading and writing data on parallel file systems. Identifying these I/O performance bottlenecks and attributing root causes are critical steps toward devising optimization strategies. Several existing studies analyze I/O logs of a set of benchmarks or applications that were run with controlled behaviors. However, there is still a lack of general approach that systematically identifies I/O performance bottlenecks for applications running "in the wild" on production systems. In this study, we have developed an analysis approach of "zooming in" from platform-wide to application-wide to job-level I/O logs for identifying I/O bottlenecks in arbitrary scientific applications. We analyze the logs collected on a Cray XC40 system in production over a two-month period. This study results in several insights for application developers to use in optimizing I/O behavior.

Wang, Teng↗
DOE OSTI2024

Excitonic Shockley-Read-Hall recombination in organic semiconductors

Trap-mediated recombination influences the performance of a wide range of electronic devices. The well-known Shockley-Read-Hall (SRH) expression for inorganic semiconductors is often invoked to describe the recombination rate in organic materials, although without a clear understanding of how its parameters relate to the underlying material properties or how it should be modified to account for the finite lifetime of exciton intermediates in, for example, the doped emissive layer of an organic light-emitting diode (OLED). Here, we formalize SRH recombination for organic semiconductors based on diffusive trapping and Langevin recombination. We show that including the exciton state suppresses the recombination rate in host-guest systems with type II energy level alignment whenever the interfacial gap between the host and guest molecular orbitals is comparable to the exciton energy. Furthermore, these results quantify the balance between bimolecular and trap-mediated recombination in doped OLED emissive layers, and indicate that devices with type II host-guest pairings can, in principle, beat the thermodynamic limit of their neat guest counterparts.

36 MATERIALS SCIENCE↗
DOE OSTI2021

VERAIO Software Management Plan

VERAIO is a set of utility codes used to provide a common set of input and outputs to the Virtual Environment for Reactor Applications (VERA). VERA is a collection of several different computer codes that all have a common input and output. This prevents the need to manage input and output from each individual code, allowing for ease of use and reducing errors associated with code operability. The VERAIO utilities include VERAIn, VERAView, and VERARun. Each of these utilities is described below. VERAIn is an input processor that reads an ASCII input file generated by users, parses the file, performs some error checking, and writes an XML file to be read by other VERA codes. The main purpose of VERAIn is to provide a common input to all of the VERA codes so users only need to learn one input. VERAIn is written in Perl and uses YAML configuration files to provide flexibility. VERAView is a graphical user interface (GUI) that reads a VERA HDF output file and allows users to visualize results. VERAView is written in Python. VERARun is a script that drives the VERA execution in a high performance computing (HPC) environment. Work performed at the code level supports the Quality Assurance Program Plan (QAPP) (VERA-QA-001), and VERA Software Quality Assurance Plan (VERA-QA-002).

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2014

Using computer-extracted image features for modeling of error-making patterns in detection of mammographic masses among radiology residents

Purpose: Mammography is the most widely accepted and utilized screening modality for early breast cancer detection. Providing high quality mammography education to radiology trainees is essential, since excellent interpretation skills are needed to ensure the highest benefit of screening mammography for patients. The authors have previously proposed a computer-aided education system based on trainee models. Those models relate human-assessed image characteristics to trainee error. In this study, the authors propose to build trainee models that utilize features automatically extracted from images using computer vision algorithms to predict likelihood of missing each mass by the trainee. This computer vision-based approach to trainee modeling will allow for automatically searching large databases of mammograms in order to identify challenging cases for each trainee. Methods: The authors’ algorithm for predicting the likelihood of missing a mass consists of three steps. First, a mammogram is segmented into air, pectoral muscle, fatty tissue, dense tissue, and mass using automated segmentation algorithms. Second, 43 features are extracted using computer vision algorithms for each abnormality identified by experts. Third, error-making models (classifiers) are applied to predict the likelihood of trainees missing the abnormality based on the extracted features. The models are developed individually for each trainee using his/her previous reading data. The authors evaluated the predictive performance of the proposed algorithm using data from a reader study in which 10 subjects (7 residents and 3 novices) and 3 experts read 100 mammographic cases. Receiver operating characteristic (ROC) methodology was applied for the evaluation. Results: The average area under the ROC curve (AUC) of the error-making models for the task of predicting which masses will be detected and which will be missed was 0.607 (95% CI,0.564-0.650). This value was statistically significantly different from 0.5 (p < 0.0001). For the 7 residents only, the AUC performance of the models was 0.590 (95% CI,0.537-0.642) and was also significantly higher than 0.5 (p = 0.0009). Therefore, generally the authors’ models were able to predict which masses were detected and which were missed better than chance. Conclusions: The authors proposed an algorithm that was able to predict which masses will be detected and which will be missed by each individual trainee. This confirms existence of error-making patterns in the detection of masses among radiology trainees. Furthermore, the proposed methodology will allow for the optimized selection of difficult cases for the trainees in an automatic and efficient manner.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2022

VERAIO Software Management Plan

VERAIO is a set of utility codes used to provide a common set of inputs and outputs to the Virtual Environment for Reactor Applications (VERA). VERA is a collection of several different computer codes that all have a common input and output. This prevents the need to manage input and output from each individual code, allowing for ease of use and reducing errors associated with code operability. The VERAIO utilities include VERAIn, VERAView, and VERARun. Each of these utilities is described below. VERAIn is an input processor that reads an ASCII input file generated by users, parses the file, performs some error checking, and writes an XML file to be read by other VERA codes. The main purpose of VERAIn is to provide a common input to all of the VERA codes, so users only need to learn one input. VERAIn is written in Perl and uses YAML configuration files to provide flexibility. VERAView is a graphical user interface (GUI) that reads a VERA hierarchical data format (HDF) output file and allows users to visualize results. VERAView is written in Python. VERARun is a script that drives the VERA execution in a high performance computing (HPC) environment. Work performed at the code level supports the quality assurance program plan (QAPP) (VERA-QA-001) and the VERA Software Quality Assurance Plan (VERA-QA-002).

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2019

DECA: scalable XHMM exome copy-number variant calling with ADAM and Apache Spark

Abstract Background XHMM is a widely used tool for copy-number variant (CNV) discovery from whole exome sequencing data but can require hours to days to run for large cohorts. A more scalable implementation would reduce the need for specialized computational resources and enable increased exploration of the configuration parameter space to obtain the best possible results. Results DECA is a horizontally scalable implementation of the XHMM algorithm using the ADAM framework and Apache Spark that incorporates novel algorithmic optimizations to eliminate unneeded computation. DECA parallelizes XHMM on both multi-core shared memory computers and large shared-nothing Spark clusters. We performed CNV discovery from the read-depth matrix in 2535 exomes in 9.3 min on a 16-core workstation (35.3× speedup vs. XHMM), 12.7 min using 10 executor cores on a Spark cluster (18.8× speedup vs. XHMM), and 9.8 min using 32 executor cores on Amazon AWS’ Elastic MapReduce. We performed CNV discovery from the original BAM files in 292 min using 640 executor cores on a Spark cluster. Conclusions We describe DECA’s performance, our algorithmic and implementation enhancements to XHMM to obtain that performance, and our lessons learned porting a complex genome analysis application to ADAM and Spark. ADAM and Apache Spark are a performant and productive platform for implementing large-scale genome analyses, but efficiently utilizing large clusters can require algorithmic optimizations and careful attention to Spark’s configuration parameters.

Linderman, Michael D. (ORCID:0000000296437148)↗
DOE OSTI2024

Performance of a second generation of the Fermilab Constant Fraction Discriminator ASIC for time-stamping long-strip AC-LGAD sensors.

We present the design and performance characterization results of the second generation of the novel Fermilab Constant Fraction Discriminator ASIC (FCFD) developed to readout AC-coupled low gain avalanche detector (AC-LGAD) sensors. This study presents the performance of the ASIC which was optimized specifically for reading out strip AC-LGAD sensors designed for ePIC experiment at EiC. Performance was measured using charge injection and particle beams with prototype AC-LGAD sensors wirebonded to the FCFD ASIC.

Apresyan, A.↗
DOE OSTI2018

SpaRC: scalable sequence clustering using Apache Spark

MOTIVATION: Whole genome shotgun based next-generation transcriptomics and metagenomics studies often generate 100-1000 GB sequence data derived from tens of thousands of different genes or microbial species. Assembly of these data sets requires tradeoffs between scalability and accuracy. Current assembly methods optimized for scalability often sacrifice accuracy and vice versa. An ideal solution would both scale and produce optimal accuracy for individual genes or genomes. RESULTS: In this work we describe an Apache Spark-based scalable sequence clustering application, SparkReadClust (SpaRC), that partitions reads based on their molecule of origin to enable downstream assembly optimization. SpaRC produces high clustering performance on transcriptomes and metagenomes from both short and long read sequencing technologies. It reaches near-linear scalability with input data size and number of compute nodes. SpaRC can run on both cloud computing and HPC environments without modification while delivering similar performance. Our findings demonstrate that SpaRC provides a scalable solution for clustering billions of reads from next-generation sequencing experiments, and Apache Spark represents a cost-effective solution with rapid development/deployment cycles for similar large-scale sequence data analysis problems. AVAILABILITY AND IMPLEMENTATION: https://bitbucket.org/berkeleylab/jgi-sparc.

59 BASIC BIOLOGICAL SCIENCES↗
NASA NTRSDate not supplied

New Features of the NEQAIR Radiation Code

The longest-lived code for predicting shock layer radiation, NEQAIR, is now in its 5th decade of service. Substantial changes to the code have been made over the previous decade, the most recent report of which was at the 5th Workshop on Radiation in High Temperature Gases in 2014, for the version referred to as NEQAIR14. This paper will review some of the improvements made to the NEQAIR code since then, which is now at v15.2. Some of these features are discussed briefly below. NEQAIR15 and subsequent versions have enabled parallel evaluation of multiple lines of sight. This is accomplished by utilizing the HDF5 file format and placing multiple lines into a single file, LOS.h5, which is used for both input and output. This approach enables straightforward parallel execution both over the number of lines of sight and the number of points per line. For large problems, runtime reduces linearly with the number of nodes deployed since each line is processed independently by a subset of MPI ranks. Three applications of the multi-line solver are discussed. The first has to do with performing loosely coupled radiation-flowfield solutions. In this case the computed absorption and emission coefficients are used to evaluate the total energy absorbed or emitted at each point, allowing evaluation of the volumetric source term in the flowfield. The second computation is for obtaining heat flux from nonuniform flows, which require integration over spherical co-ordinates. These are of particular interest for evaluating radiation on the vehicle backshell. This 3D option improves the angular integration scheme and allows adaptive line selection that together reduce the number of lines required by about an order of magnitude. The final application is for remote observation, which is essentially the 3D integration problem over a small solid angle. For all three of these computations, data can be stored in the HDF5 file which allows a NEQAIR run to be restarted when it times out, or to add atmospheric absorption or instrument scan functions. An additional level of parallelism is enabled in NEQAIR15.2 using GPU routines. The GPU parallelism has realized up to 8x speed-up when running on a single core but diminishes as CPU parallelism is increased. For running multi-line simulations, it may be easier to reserve a large number of CPU nodes than to obtain the number of GPU nodes required for similar performance. A GUI, known as NEQTPY, allows for reading and creating input files, running NEQAIR, and displaying results. A significant feature of NEQTPY is the ability to perform spectral fits to data. The fits can operate on a single line spectrum (radiance vs. wavelength) or a 3D input file with multiple columns of data. Other new features include improved constants, additional species, more detailed non-Boltzmann modelling, advanced user controls, the ability to read and calculate spectra from HITRAN datafiles, photodissociation and photoionization cross-sections. A “fast” automatic grid option may reduce the size and time of spectral calculations while still maintaining good accuracy for total heat flux.

Brett A Cruden↗
DOE OSTI2020

Time Resolution Measurements of EJ-232Q With Single- and Dual-Sided Readouts

In view of its importance to applications in high energy physics, nuclear security, and time-of-flight PET, the time resolution of 0.5% benzophenone-quenched EJ-232Q plastic scintillators of various lengths was studied using a fast waveform digitizer and fast PMTs. The scintillator sizes were 5 mm × 5 mm × 10, 25, 50, 75, or 100 mm. First, the timing measurements were performed by using a single PMT reading out one end. The degradation in time resolution with increasing scintillator length was studied. The irradiation geometry was also observed to have a non-negligible impact on the time resolution due to the variation of depth-of-interaction positions, especially for the scintillators with a high aspect ratio. Second, two PMTs were employed for a dual-sided readout. This readout together with the mean time method was demonstrated to be an effective approach to improve the timing performance since the depth-of-interaction-related effect was considerably mitigated. A time resolution of 79 ± 1 ps (full-width at half-maximum) has been achieved for the 5 × 5 × 10 mm 3 scintillator. The 5 × 5 × 100 mm 3 scintillator can still provide a time resolution as good as 172 ± 1 ps. Here, the energy range used for each was between 340 keVee and 450 keVee. In addition, the time information from individual PMTs in a dual-sided readout was able to show a time resolution almost identical to that determined with the mean-time-based technique after correcting for the depth-of-interaction variation for the 100 mm long scintillator.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2017

Partitioned key-value store with atomic memory operations

A partitioned key-value store is provided that supports atomic memory operations. A server performs a memory operation in a partitioned key-value store by receiving a request from an application for at least one atomic memory operation, the atomic memory operation comprising a memory address identifier; and, in response to the atomic memory operation, performing one or more of (i) reading a client-side memory location identified by the memory address identifier and storing one or more key-value pairs from the client-side memory location in a local key-value store of the server; and (ii) obtaining one or more key-value pairs from the local key-value store of the server and writing the obtained one or more key-value pairs into the client-side memory location identified by the memory address identifier. The server can perform functions obtained from a client-side memory location and return a result to the client using one or more of the atomic memory operations.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2017

Targeted in situ metatranscriptomics for selected taxa from mesophilic and thermophilic biogas plants

Summary Biogas production is performed anaerobically by complex microbial communities with key species driving the process. Hence, analyses of their in situ activities are crucial to understand the process. In a previous study, metagenome sequencing and subsequent genome binning for different production‐scale biogas plants ( BGP s) resulted in four genome bins of special interest, assigned to the phyla Thermotogae , Fusobacteria , Spirochaetes and Cloacimonetes , respectively, that were genetically analysed. In this study, metatranscriptome sequencing of the same BGP samples was conducted, enabling in situ transcriptional activity determination of these genome bins. For this, mapping of metatranscriptome reads on genome bin sequences was performed providing transcripts per million ( TPM ) values for each gene. This approach revealed an active sugar‐based metabolism of the Thermotogae and Spirochaetes bins and an active amino acid‐based metabolism of the Fusobacteria and Cloacimonetes bins. The data also hint at syntrophic associations of the four corresponding species with methanogenic Archaea .

Biotechnology & Applied Microbiology↗
NASA NTRS2012

I/O Performance Characterization of Lustre and NASA Applications on Pleiades

In this paper we study the performance of the Lustre file system using five scientific and engineering applications representative of NASA workload on large-scale supercomputing systems such as NASA s Pleiades. In order to facilitate the collection of Lustre performance metrics, we have developed a software tool that exports a wide variety of client and server-side metrics using SGI's Performance Co-Pilot (PCP), and generates a human readable report on key metrics at the end of a batch job. These performance metrics are (a) amount of data read and written, (b) number of files opened and closed, and (c) remote procedure call (RPC) size distribution (4 KB to 1024 KB, in powers of 2) for I/O operations. RPC size distribution measures the efficiency of the Lustre client and can pinpoint problems such as small write sizes, disk fragmentation, etc. These extracted statistics are useful in determining the I/O pattern of the application and can assist in identifying possible improvements for users applications. Information on the number of file operations enables a scientist to optimize the I/O performance of their applications. Amount of I/O data helps users choose the optimal stripe size and stripe count to enhance I/O performance. In this paper, we demonstrate the usefulness of this tool on Pleiades for five production quality NASA scientific and engineering applications. We compare the latency of read and write operations under Lustre to that with NFS by tracing system calls and signals. We also investigate the read and write policies and study the effect of page cache size on I/O operations. We examine the performance impact of Lustre stripe size and stripe count along with performance evaluation of file per process and single shared file accessed by all the processes for NASA workload using parameterized IOR benchmark.

Saini, Subhash↗
DOE OSTI2015

SU-E-J-216: A Sequence Independent Approach for Quantification of MR Image Deformations From Brachytherapy Applicators

Purpose: MRI is increasingly used as a single imaging modality for brachytherapy treatment planning. The presence of a brachytherapy applicator may cause distortions in the images, especially at higher field strengths. Our aim is to develop a procedure to quantify these distortions theoretically for any MR-sequence and to verify the estimated deformations for clinical sequences. Methods: Image distortions due to perturbation of the B0-field are proportional to the ratio of the induced frequency shift and the read-out bandwidth of the applied sequence. By reconstructing a frequency-shift map from the phase data from a multi-echo sequence, distortions can be calculated for any MR-sequence. Verification of this method for estimating distortions was performed by acquiring images with opposing read-out directions and consequently opposing distortions. The applicator shift can be determined by rigidly matching these images. Clinically, T2W-TSE-images are used for this purpose. For pre-clinical tests, EPI-sequences with narrow read-out bandwidth (19.5–47.5Hz), consequently large distortions, were added to the set of clinical MRsequences. To quantify deformations of the Utrecht Interstitial CT/MR applicator (Elekta Brachytherapy) on a Philips Ingenia 3T MRI, pre-clinical tests were performed in a phantom with the applicator in water, followed by clinical validation. Results: Deformations observed in the narrow bandwidth EPI-images were well predicted using the frequency-shift, the latter giving an overestimation up to 30%/up to 1 voxel. For clinically applied MR-sequences distortions were well below the voxel size. In patient setup distortions determined from the frequency-shift map were at sub-voxel level (<0.7mm). Using T2W-images larger distortions were found (1–2mm). This discrepancy was caused by patient movement between/during acquisition of the T2W-images with opposing read-out directions. Conclusion: Phantom experiments demonstrated the feasibility of a clinical procedure for quantification of MR-image distortions for any MR-sequence. In a clinical set-up the distortions from a Utrecht interstitial CT/MR applicator are sub-voxel level. This work was partially funded by Elekta Brachytherpy.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2025

Performance studies of the Mu2e cosmic ray veto detector

The cosmic ray veto (CRV) detector of the Mu2e experiment consists of four layers of plastic scintillation counters that surround the detector solenoid. These counters are embedded with wavelength-shifting fibers and are read out by silicon photomultipliers (SiPMs). The performance of a subset of the CRV counters was studied in a cosmic-ray test stand. Here, using data taken over a two-year period, we report the single-layer muon detection efficiency and the rate at which the light yield degrades due to the aging of the plastic scintillation counters.

Aging↗
DOE OSTI2019

Initial Assessment of Cover Degradation Features at the Mexican Hat, Utah, UMTRCA Disposal Site - 19556

The Mexican Hat, Utah, Uranium Mill Tailings Radiation Control Act (UMTRCA) Title I Disposal Site is located on the Navajo Reservation in southeast Utah. The US Nuclear Regulatory Commission licensed the site to the US Department of Energy Office of Legacy Management (LM) for custody and long-term care. LM manages long-term surveillance and maintenance activities at the site with the principal goal of protecting human health and the environment. The disposal cell was designed to encapsulate low-level radioactive mill tailings and other residual radioactive materials (RRM) in a way that minimizes the need for active maintenance and limits radon gas emanation and precipitation infiltration. The disposal cell has a multicomponent cover, consisting of a 60 cm thick (24 inch), low-permeability, bentonite-amended radon barrier, overlain by a 15 cm thick (6 inch) sandy gravel bedding/filter layer and a 30 cm thick (12 inch) riprap erosion-protection layer. Under UMTRCA, designs intended to control and contain RRM shall be effective for up to 1000 years, to the extent reasonably achievable and, in any case, for at least 200 years. In 2016, multiple subtle depressions were identified on the rock cover along the toe and lower portions of the northeast side slope of the disposal cell. Due to concerns regarding potential impacts to disposal cell performance and erosion resistance, evaluations of the depressions and cell cover design and construction were initiated. The evaluations included reviews of the disposal cell as-built drawings and supporting design calculations for the rock cover components. Also, in 2017, a weather monitoring station was installed at the site to collect continuous site-specific meteorological data, and repeatable ground-based light imaging, detection, and ranging (lidar) topographical and aerial photogrammetry surveys were implemented to monitor for potential changes in the surface depression features. The evaluations also included visual inspections of the depressions, and in 2018, small-area manual excavations of the rock cover components (test pits) were performed to observe the conditions of the cover components, scan for radioactivity, and observe evidence of erosion on the surface of the radon barrier material. Voids and incisions were discovered within the radon barrier on the north and northeast side slopes of the disposal cell. Visual observations at test pits indicate that the bedding/filter material is highly segregated and may not be in conformance with the disposal cell engineering and construction design specifications. Post-construction development of a weak to strongly cemented material has also been observed immediately below the base of the bedding layer in test pits with observed radon barrier degradation. A series of radiological surveys have been performed at the site, and radiological readings have not exceeded natural background conditions in the vicinity of the site. Thus, the site remains protective of human health and the environment. Implementation of continuous radiological monitoring at the site will be initiated in fall 2018 through the deployment of radon cups and thermoluminescent dosimeters to obtain data that provides supporting evidence that the disposal cell remains protective of human health and the environment. Evaluations to determine the cause(s) of the depression features and associated radon barrier erosion are ongoing. Possible mechanisms of cover degradation identified thus far include segregated bedding material that would allow for higher precipitation runoff velocities, resulting in erosion of the radon barrier; post-construction cementation of radon barrier or aeolian material that could negatively affect the performance of the radon barrier; cation exchange leading to the development of cementation; and aeolian deposition affecting precipitation runoff. Moreover, geotechnical sampling and testing is planned to verify spatial cover conditions relative to design specifications. The resulting information will be used to identify possible causes for the surface depressions and associated radon barrier degradation, gain insight on potential long-term performance implications along the side slopes of the disposal cell, and identify potential corrective actions. (authors)

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗
DOE OSTI2019

Can I/O Variability Be Reduced on QoS-Less HPC Storage Systems?

For a production high-performance computing (HPC) system, where storage devices are shared between multiple applications and managed in a best effort manner, I/O contention is often a major problem. In this paper, we propose a balanced messaging-based re-routing in conjunction with throttling at the middleware level. This work tackles two key challenges that have not been fully resolved in the past: whether I/O variability can be reduced on a QoS-less HPC storage system, and how to design a runtime scheduling system that can scale up to a large amount of cores. The proposed scheme uses a two-level messaging system to re-route I/O requests to a less congested storage location so that write performance is improved, while limiting the impact on read by throttling re-routing. An analytical model is derived to guide the setup of optimal throttling factor. We thoroughly analyze the virtual messaging layer overhead and explore whether the in-transit buffering is effective in managing I/O variability. Contrary to the intuition, in-transit buffer cannot completely solve the problem. It can reduce the absolute variability but not the relative variability. The proposed scheme is verified against a synthetic benchmark as well as being used by production applications.

Huang, Dan↗