A5044394335- Parallel Computing and Optimization Techniques
- Advanced Data Storage Technologies
- Cloud Computing and Resource Management
- Distributed and Parallel Computing Systems
- Scientific Computing and Data Management
- Embedded Systems Design Techniques
- Distributed systems and fault tolerance
- IoT and Edge/Fog Computing
- Advanced Mathematical Identities
- Advanced Neural Network Applications
- Statistical and Computational Modeling
- Advanced Image and Video Retrieval Techniques
- Advanced Database Systems and Queries
- Rough Sets and Fuzzy Logic
- Analog and Mixed-Signal Circuit Design
- Radiation Effects in Electronics
- Cloud Data Security Solutions
- Security and Verification in Computing
Hasso Plattner Institute
2018-2024
University of Potsdam
2016-2023
IMDEA Software
2023
National Technical University of Athens
2023
University of Toronto
2023
IBM Research - Ireland
2023
Data transfers impose a major bottleneck in heterogenous system architectures. As mitigation strategy, compute resources can be introduced places where data occurs naturally. The increased diversity of turn affects programming models and practicalities software development for near-data kernels raises the question how those made accessible to users applications.
For the implementation of data-intensive C++ applications for cache coherent Non-Uniform Memory Access (NUMA) systems, both massive parallelism and data locality have to be considered. While has been largely understood, shared memory paradigm is still deeply entrenched in mindset many software developers. Hence, aspects NUMA systems widely neglected thus far. At first sight, applying nothing approaches might seem like a viable workaround address locality. However, we argue that developers...
The domains of parallel and distributed computing have been converging continuously up to the degree that state-of-the-art server computer systems incorporate characteristics from both domains: They comprise a hierarchy enclosures, where each enclosure houses multiple processor sockets socket again contains memory controllers. A global address space cache coherency are facilitated using layers fast interconnection technologies even across enclosures. growing popularity such creates an urge...
Cloud computing offers the potential to store, manage, and process data in highly available, scalable, elastic environments. Yet, these environments still provide very limited inflexible means for customers control their data. For example, can neither specify security of inter-cloud communication bearing risk information leakage, nor comply with laws requiring be kept originating jurisdiction, sharing third parties on a fine-granular basis. This lack hinder cloud adoption that falls under...
Abstract Certain workloads such as in‐memory databases are inherently hard to scale‐out and rely on cache‐coherent scale‐up non‐uniform memory access (NUMA) systems keep up with the ever‐increasing demand for compute resources. However, many parallel programming frameworks OpenMP do not make efficient use of large NUMA they consider data locality sufficiently. In this work, we present PGASUS , a C++ framework NUMA‐aware application development that provides integrated facilities task...
This paper describes a distributed implementation of Apache Arrow that can leverage cluster-shared load-store addressable memory is hardware-coherent only within each node. The built on the ThymesisFlow prototype leverages OpenCAPI interface to create shared address space across cluster. While structures are immutable, simplifying their use in cluster memory, this creates tables and makes them accessible
Near-data accelerators play an important role in satisfying the ever growing demand for compute resources. However, efficient integration of near-data computing resources into applications, a flexible programming model and suitable abstractions on operating system level are required. This paper presents Metal FS, framework that enables users applications to orchestrate computations NVMe+FPGA device through standard shell syntax, including pipe operator. A user-space NVMe file interface...
The “HPI Future SOC Lab” is a cooperation of the Hasso Plattner Institute (HPI) and industry partners. Its mission to enable promote exchange interaction between research community partners. HPI Lab provides researchers with free charge access complete infrastructure state art hard software. This includes components, which might be too expensive for an ordinary environment, such as servers up 64 cores 2 TB main memory. offerings address particularly from but not limited areas computer...
nd Workshop on Composable Systems (COMPSYS 2023)The future of computing systems is inevitably embracing a disaggregated and composable pattern: from clusters computers to pools resources that can be dynamically combined together tailored around applications requirements.Transitioning this new paradigm requires ground-breaking research ranging hardware architectures up models abstractions at all levels the software stack.COMPSYS brings expertise researchers developers academia industry...
In response to the increasing spatial and temporal heterogeneity of memory resources their properties, we propose SMOG, a fully configurable explicitly composable benchmark suite optimized for high-resolution time-series measurements. systems with highly heterogeneous resources, or whose non-functional properties change dynamically, it is necessary understand different workloads access characteristics in order make viable placement decisions. SMOG has been developed as benchmarking tool that...
Cost models play an important role for the efficient implementation of software systems. These can be embedded in operating systems and execution environments to optimize at run time. Even though non-uniform memory access (NUMA) architectures are dominating today's server landscape, there is still a lack parallel cost that represent NUMA system sufficiently. Therefore, existing analyzed, two-step performance assessment strategy proposed incorporates low-level hardware counters as indicators....