Pasar al contenido principal

Resultados relacionados

Resultados totales (Incluyendo duplicados): 4
Encontrada(s) 1 página(s)

UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

The Marenostrum experimental exascale platform (MEEP)

  • Fell, Alexander
  • Mazure, Daniel J.
  • Garcia, Teresa C.
  • Perez, Borja
  • Teruel, Xavier
  • Wilson, Pete
  • Davis, John D.
Nascent Open Source Instruction Set Architectures such as OpenPOWER or RISC-V, allow software/hardware co-designers to fully utilize the underlying hardware, modify it or extend it based on their needs. In this paper, we introduce the vision of the MareNostrum Experimental Exascale Platform (MEEP), an Open Source platform enabling software and hardware stack experimentation targeting the High-Performance Computing (HPC) ecosystem. MEEP is built with state-of-the-art FPGAs that support PCIe and High Bandwidth Memory (HBM), making it ideal to emulate chiplet-based HPC accelerators such as ACME, at the chip, package, and/or system level. MEEP provides an FPGA Shell containing standardized interfaces (I/O and memory), enabling an emulated accelerator to communicate with the hardware of the FPGA and ensures quick integration. The first demonstration of MEEP is mapping a new accelerator, the Accelerated Compute and Memory Engine (ACME), on to this digital laboratory. This enables exploration of this novel disaggregated architecture, which separates the computation from the memory operations, optimizing the accelerator for both dense (compute-bound) as well as sparse (memory-bandwidth bound) workloads. Dense workloads focus on the computational capabilities of the engine, while dedicated processors for memory accesses optimize non-unit stride and/or random memory accesses required by sparse workloads. MEEP is an open source digital laboratory that can provide a future environment for full-stack co-design and pre-silicon exploration. MEEP invites software developers and hardware engineers to build the application, compiler, libraries and the hardware to solve future challenges in the HPC, AI, ML, and DL domains., This work has been supported by the EU H2020 project MareNostrum Experimental Exascale Platform (MEEP), and funded by the European Commission under the grant agreement No. 946002., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

Towards high scalability and fine-grained parallelism on distributed HPC platforms

  • De Haro Ruiz, Juan Miguel|||0000-0002-7427-9118
  • Álvarez Martínez, Carlos|||0000-0003-0536-5183
  • Jiménez González, Daniel|||0000-0001-6064-7883
  • Morais, Lucas Henrique
  • Martorell Bofill, Xavier|||0000-0002-0417-3430
Current High-Performance Computing systems rely on massive parallelism to achieve exascale performance. They use task scheduling and message-passing programming models to explore complementary sources of parallelism. Combining the two holds the promise of allowing seamless exploitation of both intra- and inter-node concurrency while leveraging widely-known programming abstractions. Still, the interaction between the two raises coordination problems that could make work distribution excessively costly, limiting performance. This work is the first to evaluate comprehensive hardware acceleration of their combined use integrating them in a programming model that further exploits their synergies. The hardware/software co-design approach proposed for this purpose is prototyped on a cluster of 64 FPGA nodes, where each holds a RISC-V Rocket Chip CPU with 8 cores. On one hand, this article combines OMPIF and Picos, which are hardware accelerators for message passing and task scheduling respectively. They interface with the CPU through RoCC-based custom RISC-V instructions. On the other hand, we present the Implicit Message Passing (IMP) programming model, that extends task scheduling abstractions to leverage MPI-mediated inter-node parallelism without requiring explicit MPI calls. Thus, IMP transparently allows the dataflow-style execution induced by task scheduling to span multiple nodes. We implement three benchmarks, N-body, Heat, and Cholesky, each with two different strategies, IMP and explicit MPI, and evaluate them on the multi-core FPGA-based cluster. We demonstrate our hardware-software co-design approach achieves near-linear scalability with IMP and the OMPIF/Picos accelerators, and reduces task management overhead from 2200 to 300 cycles per task. Furthermore, when leveraging all 512 cores (split among the 64 nodes), we measure speedups of 2.04x (40x in communication), 1.25x (7x in communication), and 7.29x (25x in communication) compared with unaccelerated MPI for N-body, Heat, and Cholesky respectively. Finally, at 64 nodes, we respectively achieve 99%, 83%, and 79% of weak scaling efficiency., We thank the MEEP project [Horizon 2020 grant number 946002] for providing us the FPGA cluster used in this work. This work is promoted by the Barcelona Zettascale Laboratory, backed by the Ministry for Digital Transformation and of Public Services, within the framework of the Recovery, Transformation, and Resilience Plan - funded by the European Union - NextGenerationEU. This publication is part of the research project ST4HPC - PID2023-147979NB-C21 funded by MICIU/AEI/10.13039/501100011033 and by ERDF, EU. This work is also supported by the Generalitat de Catalunya [grant number 2021 SGR 01007]., Peer Reviewed
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

Towards high scalability and fine-grained parallelism on distributed HPC platforms

  • De Haro Ruiz, Juan Miguel
  • Álvarez Martínez, Carlos
  • Jiménez González, Daniel
  • Morais, Lucas Henrique
  • Martorell Bofill, Xavier
Current High-Performance Computing systems rely on massive parallelism to achieve exascale performance. They use task scheduling and message-passing programming models to explore complementary sources of parallelism. Combining the two holds the promise of allowing seamless exploitation of both intra- and inter-node concurrency while leveraging widely-known programming abstractions. Still, the interaction between the two raises coordination problems that could make work distribution excessively costly, limiting performance. This work is the first to evaluate comprehensive hardware acceleration of their combined use integrating them in a programming model that further exploits their synergies. The hardware/software co-design approach proposed for this purpose is prototyped on a cluster of 64 FPGA nodes, where each holds a RISC-V Rocket Chip CPU with 8 cores. On one hand, this article combines OMPIF and Picos, which are hardware accelerators for message passing and task scheduling respectively. They interface with the CPU through RoCC-based custom RISC-V instructions. On the other hand, we present the Implicit Message Passing (IMP) programming model, that extends task scheduling abstractions to leverage MPI-mediated inter-node parallelism without requiring explicit MPI calls. Thus, IMP transparently allows the dataflow-style execution induced by task scheduling to span multiple nodes. We implement three benchmarks, N-body, Heat, and Cholesky, each with two different strategies, IMP and explicit MPI, and evaluate them on the multi-core FPGA-based cluster. We demonstrate our hardware-software co-design approach achieves near-linear scalability with IMP and the OMPIF/Picos accelerators, and reduces task management overhead from 2200 to 300 cycles per task. Furthermore, when leveraging all 512 cores (split among the 64 nodes), we measure speedups of 2.04x (40x in communication), 1.25x (7x in communication), and 7.29x (25x in communication) compared with unaccelerated MPI for N-body, Heat, and Cholesky respectively. Finally, at 64 nodes, we respectively achieve 99%, 83%, and 79% of weak scaling efficiency., We thank the MEEP project [Horizon 2020 grant number 946002] for providing us the FPGA cluster used in this work. This work is promoted by the Barcelona Zettascale Laboratory, backed by the Ministry for Digital Transformation and of Public Services, within the framework of the Recovery, Transformation, and Resilience Plan - funded by the European Union - NextGenerationEU. This publication is part of the research project ST4HPC - PID2023-147979NB-C21 funded by MICIU/AEI/10.13039/501100011033 and by ERDF, EU. This work is also supported by the Generalitat de Catalunya [grant number 2021 SGR 01007]., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

The Marenostrum experimental exascale platform (MEEP)

  • Fell, Alexander
  • Mazure, Daniel J.
  • Garcia, Teresa C.
  • Perez, Borja
  • Teruel, Xavier
  • Wilson, Pete
  • Davis, John D.
Nascent Open Source Instruction Set Architectures such as OpenPOWER or RISC-V, allow software/hardware co-designers to fully utilize the underlying hardware, modify it or extend it based on their needs. In this paper, we introduce the vision of the MareNostrum Experimental Exascale Platform (MEEP), an Open Source platform enabling software and hardware stack experimentation targeting the High-Performance Computing (HPC) ecosystem. MEEP is built with state-of-the-art FPGAs that support PCIe and High Bandwidth Memory (HBM), making it ideal to emulate chiplet-based HPC accelerators such as ACME, at the chip, package, and/or system level. MEEP provides an FPGA Shell containing standardized interfaces (I/O and memory), enabling an emulated accelerator to communicate with the hardware of the FPGA and ensures quick integration. The first demonstration of MEEP is mapping a new accelerator, the Accelerated Compute and Memory Engine (ACME), on to this digital laboratory. This enables exploration of this novel disaggregated architecture, which separates the computation from the memory operations, optimizing the accelerator for both dense (compute-bound) as well as sparse (memory-bandwidth bound) workloads. Dense workloads focus on the computational capabilities of the engine, while dedicated processors for memory accesses optimize non-unit stride and/or random memory accesses required by sparse workloads. MEEP is an open source digital laboratory that can provide a future environment for full-stack co-design and pre-silicon exploration. MEEP invites software developers and hardware engineers to build the application, compiler, libraries and the hardware to solve future challenges in the HPC, AI, ML, and DL domains., This work has been supported by the EU H2020 project MareNostrum Experimental Exascale Platform (MEEP), and funded by the European Commission under the grant agreement No. 946002., Peer Reviewed, Postprint (published version)