Collective Communications on a Reconfigurable Optical Interconnect
نویسندگان
چکیده
In this section, we introduce some techniques for implementing the beam routing, and collision avoidance in the recievers. Each node has a beam router, such as a computer generated hologram or a deformable mirror, which is capable of redirecting k (or 1 in the single-port model) optical beams concurrently. The reconfiguration (programming the computer generated hologram, or deforming the mirror) takes d time steps (d is an integer) while the actual transmission takes only one step. By current technology, programming a computer generated hologram takes a few microseconds [2]. Each node has a transmitter for sending an optical beam toward the beam router to be directed or distributed to the receivers at the other nodes. Also, each node has a number of receivers at its input ports. Some of these input ports may be used for collective communications while others can be used for other pair-wise communications. The number of receivers can also be degenerated to 1 if, for example, more than one operation (either collective or pair-wise communication) is supported. Path-setup phase can be done by sending an encoded light beam to the beam router to reprogram the computer generated hologram, or to deform the mirror so that the actual message can be delivered to the destination(s) directly. We propose two alternative ways for path setup phase in the circuit switching between any two nodes. A node that wants to reconfigure its optical interconnect may include the destination address in the header of its message. The router at the computer generated hologram or the deformable mirror upon receiving the message stores it in a buffer and then configures its output links so that it can forward the message to the destination node(s). This approach needs a buffer for the entire message at each beam router which is of high cost. The better approach may be to send a single beam including only the destination address to the beam router for path setup phase. Then, d steps later, the second beam containing the actual message can be sent to the already configured beam router. Several beams may arrive at a destination node at the same time. If we want to assign only one fixed wavelength to these beams, resembling the address of the receiving node (that is the transmitters are tunable, and the receivers are fixed and tuned to the wavelengths resembling their node address), then there would be a …
منابع مشابه
Collective Communications on a Reconfigurable Optical Interconnect
Free-space optical interconnection is used to fashion a reconfigurable network. Latency hiding techniques are used to enhance its performance.For this network, we present and analyze broadcasting/multi-broadcasting algorithms that utilizes latency hiding and reconfiguration. A combined total exchange algorithm has been proposed based on a combination of the direct, and standard exchange algorit...
متن کاملA novel reconfigurable optical interconnect architecture using an Opto-VLSI processor and a 4-f imaging system.
A novel reconfigurable optical interconnect architecture for on-board high-speed data transmission is proposed and experimentally demonstrated. The interconnect architecture is based on the use of an Opto-VLSI processor in conjunction with a 4-f imaging system to achieve reconfigurable chip-to-chip or board-to-board data communications. By reconfiguring the phase hologram of an Opto-VLSI proces...
متن کاملCollective Communications on a Reconfigurable Optical
Free-space optical interconnection is used to fashion a reconfigurable network. Latency hiding techniques are used to enhance its performance. For this network, we present and analyze broadcasting/multi-broadcasting algorithms that utilizes latency hiding and reconfiguration. A combined total exchange algorithm has been proposed based on a combination of the direct, and standard exchange algori...
متن کاملCommunications Latency Hiding Techniques for a Reconfigurable Optical Interconnect: Benchmark Studies
Message-passing multicomputers are composed of a large number of processor/ memory modules that communicate with each other by exchanging messages through their interconnection networks. The communication overhead is one of the most important factors affecting their performance. In this work, we present evidence (through the analysis of several parallel benchmarks) that there exists communicati...
متن کاملAll-Optical Reconfigurable-Tunable 1×N Power Splitter Using Soliton Breakup
In this paper, we numerically simulated a glass-based all-optical 1×N power splitter with eleven different configurations using soliton breakup in a nonlinear medium. It is shown that in addition to reconfigurability of the proposed splitter, its power splitting ratio is tunable up to some extent values too. Nonlinear semivectorial iterative finite difference beam propagation method (IFD-...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
عنوان ژورنال:
دوره شماره
صفحات -
تاریخ انتشار 1997