A5035287543- Advanced Data Storage Technologies
- Parallel Computing and Optimization Techniques
- Caching and Content Delivery
- Distributed and Parallel Computing Systems
- Distributed systems and fault tolerance
- Scientific Computing and Data Management
- Cloud Computing and Resource Management
- Interconnection Networks and Systems
- Algorithms and Data Compression
- Advanced Memory and Neural Computing
- Modeling, Simulation, and Optimization
- Energy Load and Power Forecasting
- Insurance and Financial Risk Management
- Privacy-Preserving Technologies in Data
- Ferroelectric and Negative Capacitance Devices
- Error Correcting Code Techniques
- Digital Rights Management and Security
- Security and Verification in Computing
- Cloud Data Security Solutions
Johannes Gutenberg University Mainz
2019-2025
Sharif University of Technology
2015-2021
Key-value stores in Cloud environments can contain more than 2 45 unique elements and be larger 100 PByte. B + -Tree s are well suited for these larger-than-memory datasets seamlessly index data stored on thousands of secondary storage devices. Unfortunately, it is often uneconomical to even store all inner tree nodes memory dataset sizes. Therefore, lookup performance affected by the additional IOs reading nodes. This number reduced increasing size leaf We propose HLN-Tree s, which support...
With the emergence of Non-Volatile Memories (NVMs) and their shortcomings such as limited endurance high power consumption in write requests, several studies have suggested hybrid memory architecture employing both Dynamic Random Access Memory (DRAM) NVM a system. By conducting comprehensive experiments, we observed that lack to consider very important aspects memories including effect of: a) data migrations on performance, b) power, c) granularity migration. This paper presents an efficient...
Emerging Non-Volatile Memories (NVMs) have promising advantages (e.g., lower idle power, higher density, and non-volatility) over the existing predominant main memory technology, DRAM. Yet, NVMs also disadvantages longer latencies, active limited endurance). System architects are therefore examining hybrid DRAM-NVM memories to enable of while avoiding as much possible. Unfortunately, design space is very large complex due existence different types their rapidly-changing characteristics....
In recent years, SSDs have gained tremendous attention in computing and storage systems due to significant performance improvement over HDDs. The cost per capacity of SSDs, however, prevents them from entirely replacing HDDs such systems. One approach effectively take advantage is use as a caching layer store critical data blocks reduce the number accesses disk subsystem. Due characteristics Flash-based limited write endurance long latency on operations, employing algorithms at Operating...
HPC application developers and administrators need to understand the complex interplay between compute clusters storage systems make effective optimization decisions. Ad hoc investigations of this based on isolated case studies can lead conclusions that are incorrect or difficult generalize. The I/O Trace Initiative aims improve scientific community's understanding operations by building a searchable collaborative archive traces from wide range applications machines, with focus...
The availability of non-volatile main memory (NVMM) has started a new era for storage systems and NVMM specific file can support extremely high data metadata rates, which are required by many HPC data-intensive applications. Scaling performance within is nevertheless often restricted the Linux kernel stack, while simply moving management to user space compromise security or flexibility.
Storage subsystem is considered as the performance bottleneck of computer systems in data-intensive applications. Solid-State Drives (SSDs) are emerging storage devices which unlike Hard Disk (HDDs), do not have mechanical parts and therefore, superior compared to HDDs. Due high cost SSDs, entirely replacing HDDs with SSDs economically justified. Additionally, can endure a limited number writes before failing. To mitigate shortcomings while taking advantage their performance, SSD caching...
In this paper, we propose an Efficient Two-Level I/O Caching Architecture (ETICA) for virtualized platforms that can significantly improve latency, endurance, and cost (in terms of cache size) while preserving the reliability write-pending data blocks. As opposed to previous one-level caching schemes in platforms, our proposed architecture 1) provides two levels by employing both Dynamic Random-Access Memory (DRAM) SSD layer 2) effectively partitions space between running VMs achieve maximum...
The Flash Translation Layer (FTL) plays a crucial role for the performance and lifetime of SSDs. It has been difficult to evaluate different FTL strategies in real SSDs past, as deeply embedded into SSD hardware. Recent host-based architectures like ZNS now enable researchers implement new strategies. In this paper, we overhead various garbage collection using host-side FTL, show their limitations when scaling size or number outstanding requests. To address these limitations, propose...
In recent years, enterprise Solid-State Drives (SSDs) are used in the caching layer of high-performance servers to close growing performance gap between processing units and storage subsystem. SSD-based I/O is typically not effective workloads with burst accesses which itself becomes bottleneck because large number accesses. Existing cache architectures mainly focus on maximizing hit ratio while they neglect average queue time Previous studies suggested bypassing when identified. These...
All-flash storage (AFS) systems have become an essential infrastructure component to support enterprise applications, where sub-millisecond latency and very high throughput are required. Nevertheless, the price per capacity ofsolid-state drives (SSDs) is relatively high, which has encouraged system architects adoptdata reduction techniques, mainlydeduplication andcompression, in solutions. To provide higher reliability performance, SSDs typically grouped usingredundant array of independent...
Checkpointing is a popular resilience method in HPC and its efficiency highly depends on the choice of checkpoint interval. Standard analytical approaches optimize intervals for big, long-running jobs that fail with high probability, while they are unable to minimize checkpointing overheads low or medium probability failing. Nevertheless, our analysis batch traces four systems shows these extremely common.We therefore propose an iterative algorithm compute efficient risk failure. The also...
Solid-State Drives (SSDs) have significant performance advantages over traditional Hard Disk (HDDs) such as lower latency and higher throughput. Significantly price per capacity limited lifetime, however, prevents designers to completely substitute HDDs by SSDs in enterprise storage systems. SSD-based caching has recently been suggested for systems benefit from of while minimizing the overall cost. While conventional algorithms Least Recently Used (LRU) provide high hit ratio processors, due...
Deduplication and compression are powerful techniques to reduce the ratio between quantity of logical data stored physical amount consumed storage. can impose significant performance overheads, as duplicate detection for large systems induces random accesses backend These have led concern that deduplication primary storage HDDs not compatible. Most inline reduction solutions therefore optimized SSDs discourage their use HDDs, even sequential workloads. In this work, we show these concerns...
Memory access efficiency is significantly enhanced by caching recent address translations in the CPUs' Translation Lookaside Buffers (TLBs). However, since operating system not aware of which core using a particular mapping, it flushes TLB entries across all cores where application runs whenever addresses are unmapped, ensuring security and consistency. These flushes, known as shootdowns, costly create performance scalability bottleneck. A key contributor to shootdowns memory-mapped I/O,...
Data reduction technologies have proven their effectiveness to decrease the ever-growing demands on storage system capacities, but also introduce new complexity in I/O stack that can easily invalidate well-known best practices. In this paper, we conduct an extensive set of experiments enterprise all-flash (AFS) equipped with open-source data module, i.e., RedHat VDO, and reveal novel observations performance gap between state-of-the-art optimal AFS integrated reduction. We then offer...
Emerging persistent memory in commodity hardware allows byte-granular accesses to state at speeds. However, prevent inconsistent due unexpected system failures, different write-semantics are required compared volatile memory. Transaction-based library solutions for facilitate the atomic modification of data languages where is explicitly managed by programmer, such as C/C++. For that provide extended capabilities like automatic management, a more native integration into language needed...
With evolving techniques to parallelize Deep Learning (DL) and the growing amount of training data model complexity, High-Performance Computing (HPC) has become increasingly important for machine learning engineers. Although many compute clusters already use accelerators or GPUs, HPC storage systems are not suitable I/O requirements DL workflows. Therefore, users typically copy whole worker nodes distribute partitions. Because depends on randomized input data, prior work stated that...
Non-volatile main memories (NVMMs) offer a comparable performance to DRAM, while requiring lower static power consumption and enabling higher densities. NVMM therefore can provide opportunities for improving both energy efficiency costs of memory. Previous hybrid memory management approaches HPC either do not consider the unique characteristics NVMMs, depend on high profiling costs, or need source code modifications. In this paper, we investigate applications' behaviors in presence as part...
Data-driven applications are becoming increasingly important in numerous industrial and scientific fields, growing the need for scalable data storage, such as object storage. Yet, many data-driven cannot use interfaces directly often have to rely on third-party file system connectors that support only a basic representation of objects files flat namespace. With sometimes millions per bucket, this simple organization is insufficient users who usually interested small subset objects. These...