TERAFLUX: Harnessing dataflow in next generation teradevices

Giorgi, Roberto; Badía, Rosa M.; Bodin, François; Cohen, Albert; Evripidou, Paraskevas; Faraboschi, Paolo; Fechner, Bernhard; Gao, Guang R.; Garbade, Arne; Gayatri, Rahulkumar; Girbal, Sylvain; Goodman, Daniel; Khan, Behram; Koliai, Souad; Landwehr, Joshua; Lê, Nhat Minh; Li, Feng; Luján, Mikel; Mendelson, Avi; Morin, Laurent; Navarro, Nacho; Patejko, Tomasz; Pop, Antoniu; Trancoso, Pedro; Ungerer, Theo; Watson, Ian; Weis, Sebastian; Zuckerman, Stéphane; Valero, Mateo

doi:10.1016/j.micpro.2014.04.001

Cited by 60 publications

(16 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…All these units exchange information with a centralized module connected to the shared bus (see figure 1). Our data-flow execution model is designed in such a way it can spawn a very high number of concurrent fine-grain threads (Data-Flow Threads -DFTs) [5,6], each of them composed of few tens of instructions. Load operations of input data are performed at the beginning of the execution, while store operations of values for other threads are performed at the end.…”

Section: System Overviewmentioning

confidence: 99%

Enhancing an x86_64 multi-core architecture with data-flow execution support

Mondelli

Scionti

et al. 2015

Proceedings of the 12th ACM International Conference on Computing Frontiers

Self Cite

View full text Add to dashboard Cite

Future exascale machines will require multi-/ many-core architectures able to efficiently run multi-threaded applications. Data-flow execution models have demonstrated to be capable of improving execution performance by limiting the synchronization overhead. This paper proposes to augment cores with a minimalistic set of hardware units and dedicated instructions that allow efficiently scheduling the execution of threads on the basis of data-flow principles. Experimental results show performance improvements of the system when compared with other techniques (e.g., OpenMP, Cilk) 1 .

show abstract

Section: System Overviewmentioning

confidence: 99%

Enhancing an x86_64 multi-core architecture with data-flow execution support

Mondelli

Scionti

et al. 2015

Proceedings of the 12th ACM International Conference on Computing Frontiers

Self Cite

View full text Add to dashboard Cite

show abstract

“…A recent project that investigated how to exploit dataflow concepts in many-cores was TERAFLUX [8], that introduced dynamic dataflow based threads called DF-threads [9]. Its challenging goal was to develop a dataflow based execution model on standard off-the-shelf cores [10] [11].…”

Section: Related Workmentioning

confidence: 99%

A matrix multiplier case study for an evaluation of a configurable dataflow-machine

Verdoscia

Vaccaro

Giorgi

2015

Proceedings of the 12th ACM International Conference on Computing Frontiers

Self Cite

View full text Add to dashboard Cite

Configurable computing has become a subject of a great deal of research given its potential to greatly accelerate a wide variety of applications that require high throughput. In this context, the dataflow approach is still promising to accelerate the kernel of applications in the field of HPC. That tanks to a computational dataflow engine able to execute dataflow program graphs directly in a custom hardware. On the other hand, evaluating radically different models of computation remains yet an open issue. In this paper we present as case study the matrix multiplication that constitutes the fundamental kernel of the linear algebra. The evaluation takes into account the execution of the matrix product both in non-pipelined and pipelined modes. Results obtained running the execution of the two modes on an FPGA-based demonstrator show the validity of the configurable Dataflow-Machine. Moreover, at the same throughput, the power consumption is expected to be lower than in clock-based systems.

show abstract

“…In the context of the TERAFLUX project [19], [20], [21] such data-flow model had been extended to multiple nodes executing seamlessly thanks to the support of an appropriate memory model [22], [14]. In such memory model a combination of consumer-producer patterns [23], [24] and transactional memory [25], [26] permits a novel combination of data-flow concepts and transactions in order to address the consistency across nodes, where each node is assumed to be cachecoherent, i.e., like in a classical multi-core.…”

Section: Related Workmentioning

confidence: 99%

Scalable Embedded Systems: Towards the Convergence of High-Performance and Embedded Computing

Giorgi

2015

2015 IEEE 13th International Conference on Embedded and Ubiquitous Computing

Self Cite

View full text Add to dashboard Cite

Embedded System toolchains are highly customized for a specific System-on-Chip (SoC).\ud When the application needs more performance, the designer is typically forced\ud to adopt a new SoC and possibly another toolchain.\ud The rationale for not scaling performance by using, e.g., two SoCs, is\ud that maintining most of the operations on-chip may allow for higher energy efficiency.\ud We are exploring the feasibility and trade-offs of designing and manufacturing a new Single Board Computer (SBC) that could serve flexibly for a number of current and future applications, by allowing scalability through clusters of SBCs while keeping the same programming model for the SBC.\ud This board is based on FPGAs and embedded processors, and its key points are: i) a fast custom interconnect for board-to-board communication and ii) an easily programmable environment which would allow both the off-loading of code into accelerators (either soft-IP blocks or hard-IP blocks) and, at the same time, the distribution of computation across boards.\ud A key challenge to successfully deploying this paradigm is to properly distribute the threads across several boards without the explicit intervention of the programmer.\ud In this paper we describe how to dynamically and efficiently distribute the computational threads in symbiosis with an appropriate memory model to allow the system scalability, so that we can double the performance by simply connecting two boards without i) changing the basic hardware components (e.g., to a different System-On-Chip) and ii) changing the programming model to follow the vendor specific toolchain. Our approach is to reduce data movement across boards. Our initial experiments have confirmed the feasibility of our approach

show abstract

TERAFLUX: Harnessing dataflow in next generation teradevices

Cited by 60 publications

References 36 publications

Enhancing an x86_64 multi-core architecture with data-flow execution support

Enhancing an x86_64 multi-core architecture with data-flow execution support

A matrix multiplier case study for an evaluation of a configurable dataflow-machine

Scalable Embedded Systems: Towards the Convergence of High-Performance and Embedded Computing

Contact Info

Product

Resources

About