Title :
Design and Implementation of a Nonblocking 2-Dilated Flattened Butterfly Switching Network
Author :
Thamarakuzhi, A. ; Chandy, J.A.
Author_Institution :
Univ. of Connecticut, Storrs, CT, USA
fDate :
7/1/2011 12:00:00 AM
Abstract :
High-performance computing is highly dependent on the communication network connecting the nodes. In this paper, we present a design and implementation of the 2-dilated flattened butterfly (2DFB) network topology, which provides nonblocking performance for relatively low cost overhead. The flattened butterfly is known to be a cost efficient topology for high-radix networks. Because of its inherent path diversity, it is able to provide comparable cost/performance with the folded-Clos network in adversarial traffic conditions. However, the flattened butterfly is a blocking network and this blocking behavior can degrade the performance of the network when all compute nodes are transmitting and receiving data at full bandwidth. 2DFB network exhibits nonblocking behavior which is a critical factor in deciding the quality of any high performance computing system network. We have implemented the network using the NetFPGA as the switching element and verified the nonblocking behavior of 2DFB. We also compare the performance of a 2DFB with other switching networks. We also show that the 2DFB topology can be used to build high speed switching systems with reduced cost.
Keywords :
field programmable gate arrays; hypercube networks; network topology; 2-dilated flattened butterfly switching network; 2DFB network; NetFPGA; blocking network; communication network; high performance computing system network; high speed switching systems; high-radix network; network topology; nonblocking performance; path diversity; Field programmable gate arrays; Hypercubes; Irrigation; Network topology; Switches; Topology; USA Councils; Aurora; NetFPGA; Nonblocking; SATA; Switching topology;
Journal_Title :
Latin America Transactions, IEEE (Revista IEEE America Latina)
DOI :
10.1109/TLA.2011.5993743