A5066266424- Advanced Data Storage Technologies
- Parallel Computing and Optimization Techniques
- Distributed and Parallel Computing Systems
- Caching and Content Delivery
- Cloud Computing and Resource Management
- Cardiomyopathy and Myosin Studies
- Distributed systems and fault tolerance
- Cardiovascular Function and Risk Factors
- Interconnection Networks and Systems
- Plant Ecology and Taxonomy Studies
- Animal Ecology and Behavior Studies
- Viral Infections and Immunology Research
- Blind Source Separation Techniques
- Congenital heart defects research
- Cardiac Imaging and Diagnostics
- Scientific Computing and Data Management
- Advanced MRI Techniques and Applications
- Williams Syndrome Research
- Peer-to-Peer Network Technologies
- Bryophyte Studies and Records
- Cardiac Structural Anomalies and Repair
Universidad de Murcia
2014-2025
Institut National des Sciences Appliquées de Rennes
2019
Institut de Recherche en Informatique et Systèmes Aléatoires
2019
Argonne National Laboratory
2019
Foundation for Research and Technology Hellas
2016
Czech Academy of Sciences, Institute of Computer Science
2015
FORTH Institute of Computer Science
2014
Background: Accurate segmentation of the left ventricular myocardium in cardiac MRI is essential for developing reliable deep learning models to diagnose non-compaction cardiomyopathy (LVNC). This work focuses on improving database used train these models, enhancing quality myocardial more precise model training. Methods: We present a semi-automatic framework that refines segmentations through three fundamental approaches: (1) combining neural network outputs with expert-driven corrections,...
Persistent key-value stores have emerged as a main component in the data access path of modern processing systems. However, they exhibit high CPU and I/O overhead. Today, due to power limitations it is important reduce overheads for processing.
Hyper-trabeculation or non-compaction in the left ventricle of myocardium (LVNC) is a recently classified form cardiomyopathy. Several methods have been proposed to quantify trabeculae accurately ventricle, but there no general agreement medical community use particular approach. In previous work, we DL-LVTQ, deep-learning approach for ventricular trabecular quantification based on U-Net CNN architecture. this extended and adapted DL-LVTQ cope with patients different particularities...
This paper presents the RAM enhanced disk cache project, REDCAP, a new of blocks which reduces read I/O time by using small portion main memory. The essential ideas behind REDCAP are to enlarge built-in drive, imitate its behavior, and take advantage read-ahead mechanism prefetching blocks. is intended be I/O- efficient, implements an activation-deactivation algorithm turns on off dynamically depending improvement achieved. has been implemented in Linux kernel. experimental results show that...
Current technology trends for efficient use of infrastructures dictate that storage converges with computation by placing devices, such as NVM-based cards and drives, in the servers themselves. With converged role interconnect among becomes more important achieving high I/O throughput. Given Ethernet is emerging dominant datacenters, it imperative to examine how reduce protocol overheads accessing remote over interconnects. In this paper we propose Tyche, a network directly on top Ethernet,...
Persistent key-value stores have emerged as a main component in the data access path of modern processing systems. However, they exhibit high CPU and I/O overhead. Nowadays, due to power limitations, it is important reduce overheads for processing. In this article, we propose Kreon , store that targets servers with flash-based storage, where overhead amplification are more significant bottlenecks compared randomness. We first observe two sources are: (a) The use compaction Log-Structured...
This paper presents the RAM enhanced disk cache project, REDCAP, a new of blocks which reduces read I/O time by using small portion main memory. The essential ideas behind REDCAP are to enlarge built-in drive, imitate its behavior, and take advantage read-ahead mechanism prefetching blocks. is intended be I/O- efficient, implements an activation-deactivation algorithm turns on off dynamically depending improvement achieved. has been implemented in Linux kernel. experimental results show that...
We present the design and implementation of both an enhanced type OSD device, OSD+ a metadata cluster based on it. OSD+s support data objects directory objects. A object stores file names attributes, supports metadata--related operations. profit features underlying systems used by storage nodes, achieving great flexibility, simplicity small overhead. By using devices, we show how can effectively be managed all servers in system, improving performance, scalability availability service. The...
Small I/O requests are important for a large number of modern workloads in the data center. Traditionally, storage systems have been able to achieve low rates small operations because hard disk drive (HDD) limitations that capable about 100–150 IOPS (I/O per second) spindle. Therefore, host CPU processing capacity and network link throughput relatively abundant providing these rates. With new device technologies, such as NAND Flash Solid State Drives (SSDs) non-volatile memory (NVM), it is...
We present a framework for simulating the performance obtained by different I/O system mechanisms and algorithms at same time, dynamically turning them on off to improve overall performance. A key element of this is design implementation virtual disk inside Linux kernel. Our creates block device which able simulate any hard drive with negligible overhead, without interfering regular requests. describe potential our proposal in REDCAP, RAM-based cache activated/deactivated according...
Accurate segmentation of the left ventricular myocardium in cardiac MRI is essential for developing reliable deep learning models to diagnose Left Ventricular Non-Compaction Cardiomyopathy (LVNC). This work focuses on improving database used train these models, enhancing quality myocardial more precise model training. We present a semi-automatic framework that refines segmentations through three fundamental approaches: 1) Combining neural network outputs with expert-driven corrections, 2)...
In order to know how different conditions influence the behavior of RAM Enhanced Disk Cache Project (RED-CAP), we have analyzed impact file system and REDCAP cache size. The results show that, for work-loads which exhibit some spatial locality application time can be reduced by more than 80% systems that split disk into block groups, while those do not use this division reduction 55%. has same performance as a traditional workloads with random or sequential access pattern. experiments also...
Accurate diagnosis of Left Ventricular Noncompaction Cardiomyopathy (LVNC) is critical for proper patient treatment but remains challenging. This work improves LVNC detection by improving left ventricle segmentation in cardiac MR images. Trabeculated indicates LVNC, automatic difficult. We present techniques to improve and evaluate their impact on diagnosis. Three main methods are introduced: (1) using full 800 × images rather than 512 512; (2) a clustering algorithm eliminate neural network...
This paper presents two mechanisms that can significantly improve the I/O performance of both hard and solid-state drives for read operations: KDSim REDCAP. is an in-kernel disk simulator provides a framework simultaneously simulating obtained by different system algorithms, dynamically turning them on off, or selecting between options policies, to overall performance. REDCAP RAM-based cache effectively enlarges built-in present in drives. By using KDSim, this activated/deactivated according...
The selection of the right I/O scheduler for a given workload can greatly improve performance system. But, this is not an easy task because several factors should be considered, and even deemed "best" change at any moment. So, we present Dynamic Automatic Disk Scheduling framework (DADS) that compares different Linux schedulers automatically dynamically selects which achieves best workload. implementation described here two by running instances disk simulator inside kernel, each one having...
Management of directories with millions files, accessed by thousands clients at the same time, is a problem recently identified in HPC environments. This paper introduces an OSD+-based technique to deal those directories. We use directory objects OSD+ devices for dynamically distributing huge among several servers. Directory work independently, achieving good performance and scalability. Experiments show that, using just 8 OSD+s Ext4, FPFS able create, stat delete more than 70,000, 120,000...
Summary OSD+s are enhanced object‐based storage devices (OSDs) able to deal with both data and metadata operations via directory objects, respectively. So far, we have focused on designing implementing efficient objects in OSD+s. This paper, however, presents our work also supporting describes how the coexistence of kinds each OSD+ is profited efficiently implement speed up some common file operations. We compare OSD+‐based Fusion Parallel File System (FPFS) Lustre OrangeFS through different...
Key-value (KV) stores based on LSM tree have become a foundational layer in the storage stack of datacenters and cloud services. Current approaches for achieving reliability availability favor reducing network traffic send to replicas only new KV pairs. As result, they perform costly compactions reorganize data both primary backup nodes, which increases device I/O CPU overhead, eventually hurts overall system performance. In this paper we describe Tebis, an efficient LSM-based store that...
Log-Structured Merge tree (LSM tree) Key-Value (KV) stores have become a foundational layer in the storage stacks of datacenter and cloud services. Current approaches for achieving reliability availability avoid replication at KV store level instead perform these operations higher layers, e.g., DB that runs on top store. The main reason is past designs replicated favor reducing network traffic increasing I/O size. Therefore, they costly compactions to reorganize data both primary backup...