Publications(January 2002 - December 2002)
- Yamin Li, Shietung Peng, and Wanming Chu, "Metacube
-- A New Interconnection Network for Large Scale Parallel
Systems'', Australian Computer Science Communications,
Vol.24, No.3, 2002, Australian Computer Society, pp29-36.
Abstract
- The hypercube has been widely used as the interconnection
network for parallel computers. However, in hypercubes,
the number of communication links for each node is a logarithmic
function of the total number of nodes. Therefore, the
hypercube is not a good candidate for an interconnection
network for a very large parallel computer that might
contain hundreds of thousands of nodes due to IC technology
and port number limitations. This paper introduces a new
interconnection network for very large parallel computers
called metacube (MC). An MC network has a 2-level cube
structure. An MC(k,m) network can connect
2m2k+k nodes
with m+k links per node, where k
is the dimension of the high-level cubes (classes) and
m is the dimension of the low-level cubes (clusters).
An MC network is a symmetric network with short diameter,
easy and efficient routing and broadcasting similar to
that of the hypercube. However, an MC network can connect
millions of nodes with up to 6 links per node. An MC(2,3)
with 5 links per node has 16,384 nodes and an MC(3,3)
with 6 links per node has 134,217,728 nodes. We describe
the MC network's structure, topological properties and
routing and broadcasting algorithms.
- Yamin Li, Shietung Peng, and Wanming Chu, "Efficient
Communication in Metacube: A New Interconnection Network'',
Proceedings of the International Symposium on Parallel Architectures,
Algorithms and Networks (I-SPAN 2002), Manila, Philippines,
May 2002, IEEE Computer Society Press, pp165-170.
Abstract
- This paper introduces a new interconnection network
for very large parallel computers called metacube (MC).
An MC network has a 2-level cube structure. An MC(k,m)
network connects 2m2k+k
nodes with m+k links per node, where k
is the dimension of a high-level cube and m is
the dimension of low-level cubes (clusters). An MC network
is a symmetric network with short diameter, easy and efficient
routing similar to that of hypercubes. However, an MC
network can connect more than one hundred of millions
of nodes with only 6 links per node. Design of efficient
routing algorithms for collective communications is the
key issue for any interconnection network. In this paper,
we also show that total exchange (all-to-all personalized
communication) can be done efficiently in metacube.
- Yamin Li and Shietung Peng, "Algorithms of Routing
and Matrix Multiplication on Dualcube'', Proceedings of
the Second International Conference on Software Engineering,
Artificial Intelligence, Networking and Parallel/Distributed
Computing (SNPD01), Nagoya Institute of Technology, Japan,
Aug., 2001, pp422-429.
Abstract - Dualcube
is an interconnection networks that has hypercube-like
structure with the capacity to hold much more nodes than
the conventional hypercube with the same number of links
per node. The motivation of using dualcube as an interconnection
network is to mitigate the problem of increasing the number
of links in the large-scale hypercube network while keeps
most of the topological properties of the hypercube network.
In this paper, we focus on the design of efficient algorithms
for routing and numerical operations on dualcube such
as prefix computation, vector-matrix and matrix-matrix
multiplications. Our results show that the routing and
the basic numerical computations can be done on dualcube
almost as fast as those on hypercube.
- Wanming Chu and Yamin Li, "Performance Evaluation of
a Multiple-Threaded Multiple-Pipelined Java Processor'',
Proceedings of the 6th World Multiconference on Systemics,
Cybernetics and Informatics (SCI 2002), Orlando, USA, July
14-18, 2002, Vol. V, Computer Science I, pp281-286.
Abstract - Executing
Java bytecodes natively by high-performance Java processors
has been becoming more attractive as network computing
gains importance. This paper proposes a multiple-threaded
multiple-pipelined Java processor architecture and presents
the design and implementation of a tracer which gathers
desirable information on the behavior of Java programs
and an architectural simulator which investigates Java
bytecode instruction/thread level parallelism and predicts
the performance of the proposed processor. We use multiple
pipelined functional units (FUs) to execute multiple bytecodes
in parallel in our processor model. The types of FUs and
the number of each type of FUs needed for executing Java
bytecodes are also investigated. Our simulation results
show that a Java processor with two issuing slots could
achieve an average 5.86 IPC (instructions per cycle) performance.
The simulator also predicts the utilization of FUs with
different processor configurations. Since the processor
configurations can be changed easily just by changing
a configuration file, this simulator and the simulation
results can be helpful for turning the processor design
decisions.
- Yamin Li, Shietung Peng, and Wanming Chu, "Fault-tolerant
Routing in Metacube'', Proceedings of the Third International
Conference on Parallel and Distributed Computing, Applications
and Technologies (PDCAT'02), Kanazawa Bunka Hall, Kanazawa,
Japan, September 2002, pp343-350.
Abstract
- A new interconnection network with low-degree for very
large parallel computers called metacube (MC) has been
introduced recently. The MC network has short diameter
similar to that of the hypercube. However, the degree
of an MC network is much lower than that of a hypercube
of the same size. More than one hundred of millions of
nodes can be connected by an MC network with up to 6 links
per node. % The MC network has 2-level cube structure.
An MC(k,m) network that connects 2m2k+k
nodes with m+k links per node has two parameters,
k and m, where k is the dimension
of the high-level cubes (classes) and m is the
dimension of the low-level cubes (clusters). % In this
paper, we give an efficient algorithm for fault-tolerant
routing in MC networks. The fault-tolerant routing problem
in MC(k,m) is solved through a special structure
in an MC network, called multi-channel cube. In order
to construct k disjoint paths for each node pair in a
multi-channel cube, an innovative technique, called signature,
is introduced.
- Yamin Li, Shietung Peng, and Wanming Chu, "Hamiltonian
Cycle Embedding for Fault Tolerance in Dual-cube'', Proceedings
of the IASTED International Conference on Networks, Parallel
and Distributed Processing, and Applications (NPDPA 2002),
Tsukuba, Japan, October 2002, pp1-6.
Abstract - The hypercube
has been widely used as the interconnection network (IN)
in parallel computers. However, the major drawback of
the hypercube is the increase in the number of communication
links for each node with the increase in the total number
of nodes in the system. A dual-cube DC(m) has m+1
links per node where m is the degree of a cluster
(m-cube), one more link is used for connecting
to a node in another cluster. The dual-cube mitigates
the problem of increasing number of links in the large-scale
hypercube network while keeps most of the topological
properties of the hypercube network. Embedding a linear
array or a ring into interconnection networks even when
faulty-links exit is an important issue for the design
of INs. In this paper, we show that a hamiltonian cycle
exists in a DC(m) with up to m-1 faulty links.
This is optimal because the degree of a DC(m) is
m+1. We also give efficient algorithms for constructing
hamiltonian cycles in DC(m).
- Yamin Li, Shietung Peng, and Wanming Chu, "From Dual-cube
to Metacube: Efficient Low-Degree Alternatives to Hypercube'',
Proceedings of the First International Symposium on Cyber
Worlds: Theory and Practice (CW 2002), November 2002, Tokyo,
Japan. IEEE Computer Society Press, pp85-94.
Abstract - The hypercube
has been widely used as the interconnection network in
parallel computers. However, when dealing with the parallel
computers of very large scale, the port limitation due
to the technology greatly forbid the use of hypercube
networks. The hypercube-based SGI Origin2000, a newly
developed multiprocessor system, tried to solve this problem
by introducing a Cray router. In this paper, we first
describe a hypercube-like network, called dual-cube, that
was motivated by the structure of Origin2000. A dual-cube
DC(m) has m+1 links per node where m
is the degree of a cluster (m-cube), one more link
is used for connecting to a node in another cluster. The
dual-cube mitigates the problem of port limitation in
the large-scale hypercube network while keeps most of
the topological properties of the hypercube network. Then,
we describe an interconnection network that extends dual-cube
into a more general network called metacube. The metacube
has a two-level cube structure with two parameters representing
the dimensions of the two-level cubes. Metacube is much
more flexible than dual-cube and can solve the port limitation
problem completely. The dual-cube and metacube networks
can be applied to SGI Origin2000 to connect large number
of processors without using Cray router.
>PAGE TOP
|
|
|