Communication Requirements for FPGA-Centric Molecular Dynamics
نویسندگان
چکیده
FPGA-centric clusters use FPGAs for both computation and communication and thereby address three fundamental problems of future High Performance Clusters: efficient use of silicon, power, and removing communication bottlenecks. In this study we report on the plausibility of using such clusters for Molecular Dynamics simulations, in particular by determining the communication requirements for such a cluster. We begin by reviewing MD on a single FPGA-based node and use that performance to determine the time budget for the communication. We then characterize the data communication characteristics for a production MD code (NAMD) in two ways: analytically and by instrumenting the code. We apply this information to clusters of various sizes and node complexity. The conclusion is that a cluster with 256 FPGAs distributed in 64 nodes is appropriately provisioned, even for modest simulations, with a bidirectional 3D torus where each link consists of 1-2 of an FPGAs serial ports. Keywords-High Performance Reconfigurable Computing; Molecular Dynamics; FPGA-Centric Compute Clusters
منابع مشابه
FPGA-Centric Clusters: Case Study with the 3D FFT
FPGA-centric clusters (FCCs) use FPGAs for both computation and communication and thereby address three fundamental problems of future High Performance Clusters: efficient use of silicon, power, and strong scaling. One aspect of FCCs studied here is that integrating reconfigurable compute, NIC, and switch components opens up a vast space of application-aware communication designs and optimizati...
متن کاملAn FPGA Implementation of Reciprocal Sums for SPME
Molecular Dynamics simulations have become an interesting target for acceleration using Field-Programmable Gate Arrays (FPGA). Still to be attempted completely in FPGA hardware is the computation of the Coulombic interactions using the Smooth Particle Mesh Ewald (SPME) algorithm. In this work, we describe the design, the implementation, and the verification effort of an FPGA compute engine, nam...
متن کاملModular Asynchronous Network-on-Chip: Application to GALS Systems Rapid Prototyping
This paper presents an innovating methodology for fast and easy design of Asynchronous Network-on-Chips (ANoCs) dedicated to GALS systems. A topologyindependent building-block approach permits to design modular, scalable and reliable ANoCs with low-power and low-complexity requirements. A crossbar generator is added to the existing design flow for fast system architecture exploration. A multi-c...
متن کاملDual Band Slotted Patch Antenna for On/Off-body Communication
Body Centric Wireless Communication (BCWC) has been topic of much research in recent years. In this manuscript, a novel low-cost microstrip patch antenna is designed for on-/off-body communication with a dual-band operation. The proposed structure exhibits a monopole-like radiation pattern at 2.45 GHz and a broadside radiation at 4.8 GHz, simultaneously. Lateral dimensions of 42 mm × 46 mm and ...
متن کاملSecure FPGA Design by Filling Unused Spaces
Nowadays there are different kinds of attacks on Field Programmable Gate Array (FPGA). As FPGAs are used in many different applications, its security becomes an important concern, especially in Internet of Things (IoT) applications. Hardware Trojan Horse (HTH) insertion is one of the major security threats that can be implemented in unused space of the FPGA. This unused space is unavoidable to ...
متن کامل