Logo des Repositoriums

Auflistung nach Autor:in "Pietrzyk, Johannes"

1 - 4 von 4
  • Zeitschriftenartikel
    Evaluating the Vector Supercomputer SX-Aurora TSUBASA as a Co-Processor for In-Memory Database Systems
    (Datenbank-Spektrum: Vol. 19, No. 3, 2019) Pietrzyk, Johannes; Habich, Dirk; Damme, Patrick; Focht, Erich; Lehner, Wolfgang
    In-memory column-store database systems are state of the art for the efficient processing of analytical workloads. In these systems, data compression as well as vectorization play an important role. Currently, the vectorized processing is done using regular SIMD (Single Instruction Multiple Data) extensions of modern processors. For example, Intel’s latest SIMD extension supports 512-bit vector registers which allows the parallel processing of 8× 64-bit values. From a database system perspective, this vectorization technique is not only very interesting for compression and decompression to reduce the computational overhead, but also for all database operators like joins, scan, as well as groupings. In contrast to these SIMD extensions, NEC Corporation has recently introduced a novel pure vector engine (supercomputer) as a co-processor called SX-Aurora TSUBASA. This vector engine features a vector length of 16.384 bits with the world’s highest bandwidth of up to 1.2 TB/s, which perfectly fits to data-intensive applications like in-memory database systems. Therefore, we describe the unique architecture and properties of this novel vector engine in this paper. Moreover, we present selected in-memory column-store-specific evaluation results to show the benefits of this vector engine compared to regular SIMD extensions. Finally, we conclude the paper with an outlook on our ongoing research activities in this direction.
  • Textdokument
    Fighting the Duplicates in Hashing: Conflict Detection-aware Vectorization of Linear Probing
    (BTW 2019, 2019) Pietrzyk, Johannes; Ungethüm, Annett; Habich, Dirk; Lehner, Wolfgang
    Hash tables are a core data structure in database systems, because they are fundamental for many database operators like hash-based join and aggregation. In recent years, the efficient vectorized implementation using SIMD (Single Instruction Multiple Data) instructions has attracted a lot of attention. Generally, all hash table implementations need to address what happens when collisions occur. In order to do that, the collisions have to be detected first. There are two types of collisions: (i) key duplicates and (ii) hash value duplicates. The second type is more complicated than the first type. In this paper, we investigate linear probing as a heavily applied hash table implementation and we present an extension of the state-of-the-art vectorized implementation with a hardware-supported duplicate or collision detection. For that, we use novel SIMD instructions which have been introduced with Intel’s SIMD instruction set extension AVX-512. As we are going to show, our approach outperforms the state-of-the-art vectorized version for the key handling, but introduces novel challenges for the value handling. We conclude the paper with some ideas how to tackle that challenge.
  • Textdokument
    First Investigations of the Vector Supercomputer SX-Aurora TSUBASA as a Co-Processor for Database Systems
    (BTW 2019 – Workshopband, 2019) Pietrzyk, Johannes; Habich, Dirk; Damme, Patrick; Lehner, Wolfgang
    The hardware landscape is currently changing from homogeneous multi-core systems towards heterogeneous systems with many different computing units, each with their own characteristics. This trend is a great opportunity for database systems to increase the overall performance if the heterogeneous resources can be utilized efficiently. Following that trend, NEC cooperation has recently introduced a novel heterogeneous hardware system called SX-Aurora TSUBASA. This novel heterogeneous system features a strong vector engine as a (co-)processor providing world’s highest memory bandwidth of 1.2TB/s per vector processor. From a database system perspective, where many operations are memory bound, this bandwidth is very interesting. Thus, we describe the unique architecture and properties of this novel heterogeneous system in this paper. Moreover, we present first database-specific evaluation results to show the benefit of this system to increase the query performance. We conclude the paper with an outlook on our ongoing research activities in this direction.
  • Konferenzbeitrag
    Working with Disaggregated Systems. What are the Challenges and Opportunities of RDMA and CXL?
    (BTW 2023, 2023) Geyer, Andreas; Ritter, Daniel; Lee, Dong Hun; Ahn, Minseon; Pietrzyk, Johannes; Krause, Alexander; Habich, Dirk; Lehner, Wolfgang
    The usage of disaggregated systems in large scale data-centers offers a lot of flexibility and easy scalability in comparison to the traditional statically configured scale-up and scaleout systems. Disaggregated architectures allow for the creation of software composable systems in order to create a virtual machine by software out of the pool of available hardware resources. In this paper, we propose a memory disaggregation classification and applicable use cases. We would be delighted to present our ideas and the memory disaggregation classification at the workshop and discuss the presented ideas. The valuable feedback of the attendees will help us to further refine our classification both in terms of preciseness and applicability.