Optical: systems and elements – Deflection using a moving element – Using a periodically moving element
Reexamination Certificate
1999-04-09
2002-06-25
Pascal, Leslie (Department: 2733)
Optical: systems and elements
Deflection using a moving element
Using a periodically moving element
C359S199200, C359S199200, C359S199200
Reexamination Certificate
active
06411418
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to wavelength division multiplexed (WDM) optical systems, and more specifically, it relates to optical systems, in which independent channels on different optical wavelengths are simultaneously broadcast to many nodes over a star coupler.
2. Description of Related Art
A key barrier to higher performance levels in massively parallel processors (MPPs) is the communication limits that exist among the individual processors, and between the processors and main memory. Such communication limits include delays in message transmission that could be reduced, e.g., by increasing the transmission bandwidth. The time delay for transmission of a large message reduces in proportion with the transmission bandwidth of the communication link transporting the data. Additional time delays between initial message transmission and reception stem from the use of information packets that are relayed many times, e.g., in a bucket-brigade fashion from node-to-node within a communication fabric. At each such node, the packet address header is read to route each message packet appropriately to its intended destination. If this occurs more than once, unnecessary latency in the delivery of the message packet is added and can stall processors waiting for the data. Performance suffers when the processors are starved of needed data. The processors cannot continue until all the required packets are received. The efficiency of parallel systems falls off as systems are scaled up to include more processors because of the above-mentioned latency and bandwidth limitations. As the system size, measured in number of processors, grows, each processor spends more time waiting for data. Such problems have been encountered by the Cray Research Torus program with three-dimensional interwoven rings, the Intel Paragon mesh program with two-dimensional rings without wrap-around, and the Convex Exemplar program where the symmetric multiprocessor (SMP) groups are on parallel rings.
Multiprocessing is of great current interest for both general high performance computing applications, massively parallel processing, and integrated sensor/processor systems. Increases in system node count, computing power per node, and/or sensor-generated data rate increase the communication required to maintain a balanced system that fully utilizes available computing power and sensor data. Traditional electronic solutions are not keeping pace with advances in processor performance and sensor complexity, and have increasing difficulty providing sufficient communication bandwidth. The trend towards shared memory (away from message passing) in multiprocessors places additional stress on inter-processor communications due to the short messages and rapid memory access associated with cache-to-cache coherence traffic.
The difficulty of providing sufficient communication resources between processor and memory elements in parallel, multiprocessor systems has led to many proposals to employ optical interconnects for improved bandwidth and latency. These proposals are driven by communication requirements anticipated from significant increases in computing power per node (1 GFLOPS per CPU near term) and system node count, and the recognition that traditional electronic interconnects will have increasing difficulty in meeting these requirements. Enhanced interconnects are required to provide sufficiently rapid access to remote, distributed memory so that available computing power is fully utilized for applications requiring tightly coupled multiprocessing. Cache-coherent, shared memory operation places additional stress on inter-element communications due to the short messages and rapid memory access associated with cache-to-cache coherence traffic.[
6
] In addition, rapid remote access can significantly improve memory requirements, and thus system cost, for certain scientific codes (e.g.: in which complex, underlying physics is represented by look-up tables), because large quantities of read-only data need not be replicated locally.
It is well known that the latency in a communications fabric can be reduced by increasing the “degree” of the network, which is the number of nodes (processors, memories or sensors) which can be accessed for communication by a given node without the necessity of intervening routing logic. A high network degree minimizes the number of times a packet header is processed en route to its destination, and thus minimizes the latency. This has led to several proposals to use fiber optic interconnects for multiprocessors, because the fiber optic media enables a broadcast architecture involving many nodes—that is, a high network degree. The typical architecture involves a broadcast architecture (embodied as a star coupler) and wavelength-selectable node transmitters. The multiple optical wavelengths in the network enable multiple, simultaneous communication transmissions involving different sets of source/destination node pairs.
The use of wavelength-division-multiplexed (WDM) optical systems (FIG.
1
), in which independent channels on different optical wavelengths are simultaneously broadcast to a large number (e.g.: hundreds) of nodes over a star coupler, is an attractive proposal for multiprocessor interconnects, offering the potential for wide-bandwidth, single-hop communications among all nodes. Each wavelength provides an independent, concurrent logical bus channel. With sufficient system wavelengths, it provides a non-blocking crossbar interconnect (output contention only), and can lead to a knockout switch (no output contention) given sufficient receiver resources. While scaling of such systems is ultimately limited by the optical power budget and bandwidth limitations of the optical transceiver technology, use of bridged WDM star couplers as multi-ported ported routers or spanning busses can enable scaling to higher node count. The large degree/fanout of such routers/busses is attractive for minimizing system diameter and global communication latency.
In previously proposed, conventional architectures of this type, in which a single pair of optical fibers is used to transport information to and from each node, there exists a fundamental tradeoff between the number of nodes on the star coupler (the network degree) and the transmission bandwidth. An information source must provide sufficient optical power to transmit to many destinations simultaneously because optical receivers will not produce error-free outputs unless they receive strong optical signals. The required optical signal strength increases with increasing bandwidth. When there are a lot of destinations, and the node degree increases, a larger amount of power is required. However, optical power cannot be increased indeterminately because of other system constraints, including the cost of high power laser transmitters, maximum device power limits, and the desire to operate with “eye-safe” laser powers in the network. These constraints on maximum transmission power will force the system to operate with lower transmission bandwidth when the number of nodes on a star coupler is increased. This is an undesirable option, which occurs in a variety of multiwavelength optical architectures based on broadcast-and-select type architectures, including those using n-to-n broadcast, n-to-n star couplers, or n-to-1 combining in the optical domain suffer from the power inefficiencies of 1
, where n is the number of nodes on the network. The hardware design is complicated as more wavelengths are required to be emitted from each node in a system.
Examples of the type of architecture described above are presented by
Charles Husbands in U. S. Pat. No., 5,446,572;
E. Arthurs et al.,
Electron. Lett.
24, 119 (1988);
K. Ghose, “Performance Potentials Of An Optical Fiber Bus Using Wavelength Division Multiplexing”,
Proc. SPIE
1849, 172-183 (1993);
M. Goodman et al, “The LAMBDANET Multiwavelength Network”, IEEE J. Sel. Areas in Communications vol. 8, no 6, pp 995-1004 (1990); and
H. Obara and Y
DeGroot Anthony J.
Deri Robert J.
Haigh Ronald E.
Pascal Leslie
The Regents of the University of California
Thompson Alan H.
Wooldridge John P.
LandOfFree
High-performance parallel processors based on star-coupled... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with High-performance parallel processors based on star-coupled..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and High-performance parallel processors based on star-coupled... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2898932