Browsing by Author "Liu, Jia"

Now showing 1 - 16 of 16
  • Thumbnail Image
    Best Management Practice. Fact Sheet 9, Bioretention
    Sample, David J.; Liu, Jia (Virginia Cooperative Extension, 2013-09-06)
    Explains what bioretention is, where and how it is used, its limitations, maintenance, expected costs, and a glossary of terms.
  • Thumbnail Image
    Contrasting effects of genotype and root size on the fungal and bacterial communities associated with apple rootstocks
    Liu, Jia; Abdelfattah, Ahmed; Wasserman, Birgit; Wisniewski, Michael; Droby, Samir; Fazio, Gennaro; Mazzola, Mark; Wu, Xuehong (Oxford University Press, 2022-01-05)
    The endophytic microbiome of plants is believed to have a significant impact on its physiology and disease resistance, however, the role of host genotype in determining the composition of the endophytic microbiome of apple root systems remains an open question that has important implications for defining breeding objectives. In the current study, the bacterial and fungal microbiota associated with four different apple rootstocks planted in April, 2018 in the same soil environment and harvested in May, 2019 were evaluated to determine the role of genotype on the composition of both the bacterial and fungal communities. Results demonstrated a clear impact of genotype and root size on microbial composition and diversity. The fungal community was more affected by plant genotype whereas the bacterial community was shaped by root size. Fungal and bacterial abundance was equal between different-sized roots however, significantly higher microbial counts were detected in rhizosphere samples compared to root endosphere samples. This study provides information that can be used to develop a comprehensive and readily applicable understanding of the impact of genotype and environmental factors on the establishment of plant microbiome, as well as its potential function and impact on host physiology.
  • Thumbnail Image
    Development of a Design-Based Computational Model of Bioretention Systems
    Liu, Jia (Virginia Tech, 2013-12-03)
    Multiple problems caused by urban runoff have emerged as a consequence to the continuing development of urban areas in recent decades. The increase of impervious land areas can significantly alter watershed hydrology and water quality. Typical impacts to downstream hydrologic regimes include higher peak flows and runoff volumes, shorter concentration times, and reduced infiltration. Urban runoff increases the transport of pollutants and nutrients and thus degrades water bodies adjacent to urban areas. One of the most frequently used practices to restore the hydrology and water quality of urban watersheds is bioretention (also known as a rain garden). Despite its wide applicability, an understanding of its multiple physiochemical and biological treatment processes remains an active research area. To provide a wide ability to evaluate the hydrologic input to bioretention systems, spatial and temporal distribution of storm events in Virginia were studied. Results generated from long-term frequency analysis of 60-year precipitation data demonstrate that the 90 percentile, or 10-year return period rainfall depth and dry duration in Virginia are between 22.9 – 35.6 mm and 15.3 – 25.8 days, respectively. Monte-Carlo simulations demonstrated that sampling programs applied in different regions would likely encounter more than 30% of precipitation events less than 2.54 mm, and 10% over 25.4 mm. Further experimental research was conducted to evaluate bioretention recipes for retaining stormwater nitrogen (N) and phosphorus (P). A mesocosm experiment was performed to simulate bioretention facilities with 3 different bioretention blends as media layers with underdrain pipes for leachate collection. A control group with 3 duplicates for each media was compared with a replicated vegetated group. Field measurement of dissolved oxygen (DO), oxidation-reduction potential (ORP), pH, and total dissolved solids (TDS) was combined with laboratory analyses of total suspended solids (TSS), nitrate (NO3), ammonium (NH4), phosphate (PO4), total Kjeldahl nitrogen (TKN) and total phosphorus (TP) to evaluate the nutrient removal efficacies of these blends. Physicochemical measurements for property parameters were performed to determine characteristics of blends. Isotherm experiments to examine P adsorption were also conducted to provide supplementary data for evaluation of bioretention media blends. The results show that the blend with water treatment residuals (WTR) removed >90% P from influent, and its effluent had the least TDS / TSS. Another blend with mulch-free compost retained the most (50 – 75%) total nitrogen (TN), and had the smallest DO / ORP values, which appears to promote denitrification under anaerobic conditions. Increase of hydraulic retention time (HRT) to 6 h could influence DO, ORP, TKN, and TN positively. Plant health should also be considered as part of a compromise mix that sustains vegetation. Two-way analysis of variance (ANOVA) found that single and interaction effects of HRT and plants existed, and could affect water quality parameters of mesocosm leachate. Based upon the understanding of the physiochemical and hydrologic conditions mentioned previously, a design model of a bioretention system became the next logical step. The computational model was developed within the Matlab® programming environment to describe the hydraulic performance and nutrient removal of a bioretention system. The model comprises a main function and multiple subroutines for hydraulics and treatment computations. Evapotranspiration (ET), inflow, infiltration, and outflow were calculated for hydrologic quantitation. Biomass accumulation, nitrogen cycle and phosphorus fate within bioretention systems were also computed on basis of the hydrologic outputs. The model was calibrated with the observed flow and water quality data from a field-scale bioretention in Blacksburg, VA. The calibrated model is capable of providing quantitative estimates on flow pattern and nutrient removal that agree with the observed data. Sensitivity analyses determined the major factors affecting discharge were: watershed width and roughness for inflow; pipe head and diameter for outflow. Nutrient concentrations in inflow are very influential to outflow quality. A long-term simulation demonstrates that the model can be used to estimate bioretention performance and evaluate its impact on the surrounding environment. This research advances the current understanding of bioretention systems in a systematic way, from hydrologic behavior, monitoring, design criteria, physiochemical performance, and computational modeling. The computational model, combined with the results from precipitation frequency analysis and evaluation of bioretention blends, can be used to improve the operation, maintenance, and design of bioretention facilities in practical applications.
  • Thumbnail Image
    Effect of Oligogalacturonides on Seed Germination and Disease Resistance of Sugar Beet Seedling and Root
    Zhao, Can; Wu, Chunyan; Li, Kuikui; Kennedy, John F.; Wisniewski, Michael; Gao, Lihong; Han, Chenggui; Liu, Jia; Yin, Heng; Wu, Xuehong (MDPI, 2022-07)
    Oligogalacturonides (OGs) are a bioactive carbohydrate derived from homogalacturonan. The OGs synthesized in this study significantly inhibited the mycelial growth of Rhizoctonia solani AG-4HGI in vitro, even at a low concentration (10 mg/L). The seed vigor test demonstrated that the application of 50 mg/L OGs to sugar beet seeds significantly increased average germination percentage, germination energy, germination index, and seedling vigor index. The same concentration of OGs also improved the seedling emergence percentage of sugar beet when seeds were sown in soil inoculated with D2 and D31 isolates, respectively. The lesion diameter on mature sugar beet roots caused by R. solani AG-4HGI isolates D2 and D31 also decreased by 40.60% and 39.86%, respectively, in sugar beets roots first treated with 50 mg/mL OGs in the wound site, relative to lesion size in untreated/pathogen inoculated wounds. Sugar beet roots treated with 50 mg/mL OGs prior to inoculation with the D2 isolate exhibited up-regulation of the defense-related genes glutathione peroxidase (GPX) and superoxide dismutase (SOD) by 2.4- and 1.6-fold, respectively, relative to control roots. Sugar beet roots treated with 50 mg/mL OGs prior to inoculation with D31 exhibited a 2.0- and 1.6-fold up-regulation of GPX and SOD, respectively, relative to the control. Our results indicate that OGs have the potential to be used for the protection of sugar beet against R. solani AG-4HGI.
  • Thumbnail Image
    Effect of Washing, Waxing and Low-Temperature Storage on the Postharvest Microbiome of Apple
    Abdelfattah, Ahmed; Whitehead, Susan R.; Macarisin, Dumitru; Liu, Jia; Burchard, Erik; Freilich, Shiri; Dardick, Christopher; Droby, Samir; Wisniewski, Michael (MDPI, 2020-06-23)
    There is growing recognition of the role that the microbiome plays in the health and physiology of many plant species. However, considerably less research has been conducted on the postharvest microbiome of produce and the impact that postharvest processing may have on its composition. Here, amplicon sequencing was used to study the effect of washing, waxing, and low-temperature storage at 2 °C for six months on the bacterial and fungal communities of apple calyx-end, stem-end, and peel tissues. The results of the present work reveal that tissue-type is the main factor defining fungal and bacterial diversity and community composition on apple fruit. Both postharvest treatments and low temperature storage had a strong impact on the fungal and bacterial diversity and community composition of these tissue types. Distinct spatial and temporal changes in the composition and diversity of the microbiota were observed in response to various postharvest management practices. The greatest impact was attributed to sanitation practices with major differences among unwashed, washed and washed-waxed apples. The magnitude of the differences, however, was tissue-specific, with the greatest impact occurring on peel tissues. Temporally, the largest shift occurred during the first two months of low-temperature storage, although fungi were more affected by storage time than bacteria. In general, fungi and bacteria were impacted equally by sanitation practices, especially the epiphytic microflora of peel tissues. This research provides a foundation for understanding the impact of postharvest management practices on the microbiome of apple and its potential subsequent effects on postharvest disease management and food safety.
  • Thumbnail Image
    Efficacy of the biocontrol agent Trichoderma hamatum against Lasiodiplodia theobromae on macadamia
    Li, Xiaojiao; Leng, Jinsong; Yu, Longfeng; Bai, Haidong; Li, Xiaojun; Wisniewski, Michael; Liu, Jia; Sui, Yuan (Frontiers, 2022-08)
    Macadamia (Macadamia integrifolia) trees are an important source of revenue in rainforest ecosystems. Their nuts are rich in vitamins, minerals, fiber, antioxidants, and monounsaturated oils. The fungus Lasiodiplodia theobromae, however, is a major disease problem, causing kernel rot and other disease symptoms. In the present study, a dual confrontation assay was used to evaluate the inhibitory effect of an endophytic strain of Trichoderma hamatum C9 from macadamia root against L. theobromae. Volatiles and cell-free culture filtrate of T. hamatum were also used to assess their antifungal activity against L. theobromae. Results suggested that T. hamatum exhibited a significant inhibitory effect against L. theobromae in vitro. Further results of a biocontrol assay indicated that a spray treatment of T. hamatum conidial suspension significantly decreased the size of lesions caused by artificially inoculated L. theobromae on macadamia leaves, as well as the disease index in young trees inoculated with L. theobromae, relative to sterile water controls. Collectively, our findings indicate that T. hamatum C9 represents a potential biocontrol agent that can be used to manage L. theobromae on macadamia.
  • Thumbnail Image
    Evidence for host-microbiome co-evolution in apple
    Abdelfattah, Ahmed; Tack, Ayco J. M.; Wasserman, Birgit; Liu, Jia; Berg, Gabriele; Norelli, John; Droby, Samir; Wisniewski, Michael (Wiley, 2022-06)
    Plants evolved in association with a diverse community of microorganisms. The effect of plant phylogeny and domestication on host-microbiome co-evolutionary dynamics are poorly understood. Here we examined the effect of domestication and plant lineage on the composition of the endophytic microbiome of 11 Malus species, representing three major groups: domesticated apple (M. domestica), wild apple progenitors, and wild Malus species. The endophytic community of M. domestica and its wild progenitors showed higher microbial diversity and abundance than wild Malus species. Heirloom and modern cultivars harbored a distinct community composition, though the difference was not significant. A community-wide Bayesian model revealed that the endophytic microbiome of domesticated apple is an admixture of its wild progenitors, with clear evidence for microbiome introgression, especially for the bacterial community. We observed a significant correlation between the evolutionary distance of Malus species and their microbiome. This study supports co-evolution between Malus species and their microbiome during domestication. This finding has major implications for future breeding programs and our understanding of the evolution of plants and their microbiomes.
  • Thumbnail Image
    Fox Hunting in Wild Apples: Searching for Novel Genes in Malus Sieversii
    Wisniewski, Michael; Artlip, Timothy; Liu, Jia; Ma, Jing; Burchard, Erik; Norelli, John; Dardick, Christopher (MDPI, 2020-12-14)
    Malus sieversii is considered the progenitor of modern apple (Malus pumila) cultivars and to represent a valuable source of genetic diversity. Despite the importance of M. sieversii as a source of disease resistance, stress tolerance, and novel fruit traits, little is known about gene function and diversity in M. sieversii. Notably, a publicly annotated genome sequence for this species is not available. In the current study, the FOX (Full-length cDNA OvereXpressing) gene hunting system was used to construct a library of transgenic lines of Arabidopsis in which each transgenic line overexpresses a full-length gene obtained from a cDNA library of the PI619283 accession of M. sieversii. The cDNA library was constructed from mRNA obtained from bark tissues collected in late fall–early winter, a time at which many abiotic stress-adaptative genes are expressed. Over 4000 apple FOX Arabidopsis lines have been established from the pool of transgenic seeds and cDNA inserts corresponding to various Gene Ontology (GO) categories have been identified. A total of 160 inserts appear to be novel, with no or limited homology to M. pumila, Arabidopsis, or poplar. Over 1300 lines have also been screened for freezing resistance. The constructed library of transgenic lines provides a valuable genetic resource for exploring gene function and diversity in Malus sieversii. Notably, no such library of t-DNA lines currently exists for any Malus species.
  • Thumbnail Image
    Global analysis of the apple fruit microbiome: are all apples the same?
    Abdelfattah, Ahmed; Freilich, Shiri; Bartuv, Rotem; Zhimo, V. Yeka; Kumar, Ajay; Biasi, Antonio; Salim, Shoshana; Feygenberg, Oleg; Burchard, Erik; Dardick, Christopher; Liu, Jia; Khan, Awais; Ellouze, Walid; Ali, Shawkat; Spadaro, Davide; Torres, Rosario; Teixido, Neus; Ozkaya, Okan; Buehlmann, Andreas; Vero, Silvana; Mondino, Pedro; Berg, Gabriele; Wisniewski, Michael; Droby, Samir (2021-03-18)
    We present the first worldwide study on the apple (Malus x domestica) fruit microbiome that examines questions regarding the composition and the assembly of microbial communities on and in apple fruit. Results revealed that the composition and structure of the fungal and bacterial communities associated with apple fruit vary and are highly dependent on geographical location. The study also confirmed that the spatial variation in the fungal and bacterial composition of different fruit tissues exists at a global level. Fungal diversity varied significantly in fruit harvested in different geographical locations and suggests a potential link between location and the type and rate of postharvest diseases that develop in each country. The global core microbiome of apple fruit was represented by several beneficial microbial taxa and accounted for a large fraction of the fruit microbial community. The study provides foundational information about the apple fruit microbiome that can be utilized for the development of novel approaches for the management of fruit quality and safety, as well as for reducing losses due to the establishment and proliferation of postharvest pathogens. It also lays the groundwork for studying the complex microbial interactions that occur on apple fruit surfaces.
  • Thumbnail Image
    Heterogeneous Sensor Data based Online Quality Assurance for Advanced Manufacturing using Spatiotemporal Modeling
    Liu, Jia (Virginia Tech, 2017-08-21)
    Online quality assurance is crucial for elevating product quality and boosting process productivity in advanced manufacturing. However, the inherent complexity of advanced manufacturing, including nonlinear process dynamics, multiple process attributes, and low signal/noise ratio, poses severe challenges for both maintaining stable process operations and establishing efficacious online quality assurance schemes. To address these challenges, four different advanced manufacturing processes, namely, fused filament fabrication (FFF), binder jetting, chemical mechanical planarization (CMP), and the slicing process in wafer production, are investigated in this dissertation for applications of online quality assurance, with utilization of various sensors, such as thermocouples, infrared temperature sensors, accelerometers, etc. The overarching goal of this dissertation is to develop innovative integrated methodologies tailored for these individual manufacturing processes but addressing their common challenges to achieve satisfying performance in online quality assurance based on heterogeneous sensor data. Specifically, three new methodologies are created and validated using actual sensor data, namely, (1) Real-time process monitoring methods using Dirichlet process (DP) mixture model for timely detection of process changes and identification of different process states for FFF and CMP. The proposed methodology is capable of tackling non-Gaussian data from heterogeneous sensors in these advanced manufacturing processes for successful online quality assurance. (2) Spatial Dirichlet process (SDP) for modeling complex multimodal wafer thickness profiles and exploring their clustering effects. The SDP-based statistical control scheme can effectively detect out-of-control wafers and achieve wafer thickness quality assurance for the slicing process with high accuracy. (3) Augmented spatiotemporal log Gaussian Cox process (AST-LGCP) quantifying the spatiotemporal evolution of porosity in binder jetting parts, capable of predicting high-risk areas on consecutive layers. This work fills the long-standing research gap of lacking rigorous layer-wise porosity quantification for parts made by additive manufacturing (AM), and provides the basis for facilitating corrective actions for product quality improvements in a prognostic way. These developed methodologies surmount some common challenges of advanced manufacturing which paralyze traditional methods in online quality assurance, and embody key components for implementing effective online quality assurance with various sensor data. There is a promising potential to extend them to other manufacturing processes in the future.
  • Thumbnail Image
    MIMO Wireless Networks: Modeling and Optimization
    Liu, Jia (Virginia Tech, 2010-02-04)
    A critical factor affecting the future prospects of wireless networks for wide-scale deployment is network capacity: the end users wish to have their communication experience over wireless networks to be comparable or similar to that for wireline networks. An effective approach to increase network capacity is to increase spectrum efficiency. Such an approach can be achieved by the use of multiple antenna systems (also known as multiple-input multiple-output (MIMO) technology). The benefits of substantial improvements in capacity at no cost of additional spectrum and power have positioned MIMO as one of the breakthrough technologies in modern wireless communications. As expected, research activities on applying MIMO to a variety of wireless networks have soared in recent years. However, compared with the simple point-to-point MIMO channel, which is relatively well-understood nowadays, network design and performance optimization for MIMO-based wireless networks is considerably more challenging. Many fundamental problems remain unsolved. Due to the complex characteristics of MIMO physical layer technology, it is not only desirable but also necessary to consider models and constraints at multiple layers (e.g., physical, link, and network) jointly. The formulations of these cross-layer problems for MIMO wireless networks, however, are usually mathematically challenging. In this dissertation, we aim to develop some novel algorithmic design and optimization techniques that provide optimal or near-optimal solutions. Based on network structure, this dissertation is organized into two parts. In the first part, we focus on single-hop MIMO wireless networks, while in the second part, we focus on multi-hop MIMO networks. The main results and contributions of this dissertation are summarized as follows. Single-hop MIMO Networks. In the first part of this dissertation, we study three different optimization problems for single-hop MIMO networks. The first problem addresses weighted proportional fair (WPF) scheduling associated with MIMO broadcast channels (Chapter 2). For the WPF scheduling problem in MIMO broadcast channels, we develop two algorithms that can efficiently determine the optimal dirty paper encoding order and power allocation to achieve an optimal WPF performance. To our knowledge, our work is the first that provides solutions to the WPF scheduling problem in MIMO broadcast channels. Our next problem concerns single-hop MIMO ad hoc networks (Chapter 3), which are quite different from the MIMO broadcast channels studied in the previous chapter. Single-hop MIMO ad hoc networks can be simply described as "multiple one-to-one," as compared with MIMO broadcast channels, which are "one-to-many." Performance optimization for such networks is known to be challenging due to the non-convex mathematical structure. Indeed, these networks can be viewed as the general case of interference channels in network information theory context, for which the capacity region remains unknown even under the two-user case. In this chapter, we treat the co-channel interference in the network as noise. We consider the maximum weighted sum rate problem under the single-carrier setting. We propose a global optimization approach that combines branch-and-bound (BB) and the reformulation-linearization technique (RLT). This technique is guaranteed to find a global optimal solution Multi-hop MIMO Networks. In addition to managing resources such as power and scheduling in single-hop networks, routing and end-to-end session rate control need to be considered in multi-hop MIMO networks. Thus, performance optimization problems in multi-hop MIMO networks are more interesting and yet challenging. In Chapter 4, we first consider the problem of jointly optimizing power and bandwidth allocation at each node and multihop/multipath routing in a multi-hop MIMO network that employs orthogonal channels. We show that this problem has some special structure that admits a decomposition into a set of subproblems in its dual domain. Based on this finding, we propose both centralized and distributed optimization algorithms to solve this problem optimally. In Chapter 5, we relax the orthogonal channel assumption. More specifically, we exploit the advantage of "dirty paper coding" (DPC) to allow multiple links originated from the same node to share the same channel media simultaneously. However, the formulation of cross-layer optimization problem with DPC has a non-convex structure and an exponentially large search space inherent in enumerating DPC's encoding orders. To address these difficulties, we propose an approach to reformulate and convexify the original problem. Based on the reformulated problem, we design an efficient solution procedure by exploiting decomposable dual structure. One thing in common in Chapters 4 and 5 is that we adopt the classical matrix-based MIMO channel models at the physical layer. Although this approach has its merit, the complex matrix operations in the classical MIMO models may pose a barrier for researchers in networking research community to gain fundamental understanding on MIMO networks. To bridge this gap between communications and networking communities, in Chapter 6, we propose a simple, accurate, and tractable model to enable the networking community to carry out cross-layer research for multi-hop MIMO networks. At the physical layer, we develop an accurate and simple model for MIMO channel capacity computation that captures the essence of spatial multiplexing and transmit power limit without involving complex matrix operations and the water-filling algorithm. At the link layer, we devise a space-time scheduling scheme called order-based interference cancellation (OBIC) that significantly advances the existing zero-forcing beamforming (ZFBF) to handle interference in a multi-hop network setting. The proposed OBIC scheme employs simple algebraic computation on matrix dimensions to simplify ZFBF in a multi-hop network. Finally, we apply both the new physical and link layer models to study a cross-layer optimization problem for a multi-hop MIMO network.
  • Thumbnail Image
    On Scheduling Ring-All-Reduce Learning Jobs in Multi-Tenant GPU Clusters with Communication Contention
    Yu, Menglu; Ji, Bo; Rajan, Hridesh; Liu, Jia (ACM, 2022-10-03)
    Powered by advances in deep learning (DL) techniques, machine learning and artificial intelligence have achieved astonishing successes. However, the rapidly growing needs for DL also led to communication- and resource-intensive distributed training jobs for large-scale DL training, which are typically deployed over GPU clusters. To sustain the ever-increasing demand for DL training, the so-called “ring-all-reduce” (RAR) technologies have recently emerged as a favorable computing architecture to efficiently process network communication and computation load in GPU clusters. The most salient feature of RAR is that it removes the need for dedicated parameter servers, thus alleviating the potential communication bottleneck. However, when multiple RAR-based DL training jobs are deployed over GPU clusters, communication bottlenecks could still occur due to contentions between DL training jobs. So far, there remains a lack of theoretical understanding on how to design contention-aware resource scheduling algorithms for RAR-based DL training jobs, which motivates us to fill this gap in this work. Our main contributions are three-fold: i) We develop a new analytical model that characterizes both communication overhead related to the worker distribution of the job and communication contention related to the co-location of different jobs; ii) Based on the proposed analytical model, we formulate the problem as a non-convex integer program to minimize the makespan of all RAR-based DL training jobs. To address the unique structure in this problem that is not amenable for optimization algorithm design, we reformulate the problem into an integer linear program that enables provable approximation algorithm design called SJF-BCO (Smallest Job First with Balanced Contention and Overhead); and iii) We conduct extensive experiments to show the superiority of SJFBCO over existing schedulers. Collectively, our results contribute to the state-of-the-art of distributed GPU system optimization and algorithm design.
  • Thumbnail Image
    Review and Research Needs of Bioretention Used for the Treatment of Urban Stormwater
    Liu, Jia; Sample, David J.; Bell, Cameron; Guan, Yuntao (MDPI, 2014-04-24)
    The continued development of urban areas in recent decades has caused multiple issues affecting the sustainability of urban drainage systems. The increase of impervious surface areas in urban regions alters watershed hydrology and water quality. Typical impacts to downstream hydrologic regimes include higher peak flows and runoff volumes, shorter lag times, and reduced infiltration and base flow. Urban runoff increases the transport of pollutants and nutrients and thus degrades water bodies downstream from urban areas. One of the most frequently used practices to mitigate these impacts is bioretention. Despite its widespread use, research on bioretention systems remains active, particularly in terms of mix design and nitrogen treatment. Recent research focusing on bioretention is reviewed herein. The use of mesocosms provides the ability to isolate particular treatment processes and replicate variability. Computational models have been adapted and applied to simulate bioretention, offering potential improvements to their operation, maintenance, and design. Maintenance practices are important for sustained operation and have also been reviewed. Predicting maintenance is essential to assessing lifecycle costs. Within these research areas, gaps are explored, and recommendations made for future work.
  • Thumbnail Image
    Toward Efficient Online Scheduling for Distributed Machine Learning Systems
    Yu, Menglu; Liu, Jia; Wu, Chuan; Ji, Bo; Bentley, Elizabeth (IEEE, 2021-08-13)
    Recent years have witnessed a rapid growth of distributed machine learning (ML) frameworks, which exploit the massive parallelism of computing clusters to expedite ML training. However, the proliferation of distributed ML frameworks also introduces many unique technical challenges in computing system design and optimization. In a networked computing cluster that supports a large number of training jobs, a key question is how to design efficient scheduling algorithms to allocate workers and parameter servers across different machines to minimize the overall training time. Toward this end, in this paper, we develop an online scheduling algorithm that jointly optimizes resource allocation and locality decisions. Our main contributions are three-fold: i) We develop a new analytical model that considers both resource allocation and locality; ii) Based on an equivalent reformulation and observations on the worker-parameter server locality configurations, we transform the problem into a mixed packing and covering integer program, which enables approximation algorithm design; iii) We propose a meticulously designed approximation algorithm based on randomized rounding and rigorously analyze its performance. Collectively, our results contribute to the state of the art of distributed ML system optimization and algorithm design.
  • Thumbnail Image
    Urban Stormwater: Terms and Definitions
    Sample, David J.; Barlow, Stefani; Doumar, Lia; Liu, Jia; Wang, Chih-Yu (Virginia Cooperative Extension, 2011-11-17)
    This publication is intended to provide a summary of common terms and definitions used in urban stormwater management. A companion glossary is included with each fact sheet; this document provides a compilation of the terms used in the series.
  • Thumbnail Image
    Water supply and runoff capture reliability curves for hypothetical rainwater harvesting systems for locations across the U.S. for historical and projected climate conditions
    Alamdari, Nasrin; Sample, David J.; Liu, Jia; Ross, Andrew C. (Elsevier, 2018-03-11)
    The data presented in this article are related to the research article entitled “Assessing climate change impacts on the reliability of rainwater harvesting systems” (Alamdari et al., 2018) [1]. This article evaluated the water supply and runoff capture reliability of rainwater harvesting (RWH) systems for locations across the U.S. for historical and projected climate conditions. Hypothetical RWH systems with varying storage volumes, rooftop catchment areas, irrigated areas, and indoor wSater demand based upon population from selected locations were simulated for historical (1971–1998) and projected (2041–2068) periods, the latter dataset was developed using dynamic downscaling of North American Regional Climate Change (CC) Assessment Program (NARCCAP). A computational model, the Rainwater Analysis and Simulation Program (RASP), was used to compute RWH performance with respect to the reliability of water supply and runoff capture. The reliability of water supply was defined as the proportion of demands that are met; and the reliability of runoff capture was defined as the amount stored and reused, but not spilled. A series of contour plots using the four design variables and the reliability metrics were developed for historical and projected conditions. Frequency analysis was also used to characterize the long-term behavior of rainfall and dry duration at each location. The full data set is made publicly available to enable critical or extended analysis of this work.