WO2011071499A1 - Systèmes et procédés d'arbitrage par jetons équitable - Google Patents

Systèmes et procédés d'arbitrage par jetons équitable Download PDF

Info

Publication number
WO2011071499A1
WO2011071499A1 PCT/US2009/067574 US2009067574W WO2011071499A1 WO 2011071499 A1 WO2011071499 A1 WO 2011071499A1 US 2009067574 W US2009067574 W US 2009067574W WO 2011071499 A1 WO2011071499 A1 WO 2011071499A1
Authority
WO
WIPO (PCT)
Prior art keywords
waveguide
node
hungry
arbitration
light
Prior art date
Application number
PCT/US2009/067574
Other languages
English (en)
Inventor
Nathan Lorenzo Binkert
Robert Samuel Schreiber
Original Assignee
Hewlett-Packard Development Company, L.P.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Hewlett-Packard Development Company, L.P. filed Critical Hewlett-Packard Development Company, L.P.
Priority to EP09852137.0A priority Critical patent/EP2510448B1/fr
Priority to PCT/US2009/067574 priority patent/WO2011071499A1/fr
Priority to US13/386,382 priority patent/US8564867B2/en
Priority to CN200980161282.0A priority patent/CN102483727B/zh
Priority to TW099143002A priority patent/TWI430103B/zh
Publication of WO2011071499A1 publication Critical patent/WO2011071499A1/fr

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L12/00—Data switching networks
    • H04L12/28—Data switching networks characterised by path configuration, e.g. LAN [Local Area Networks] or WAN [Wide Area Networks]
    • H04L12/42—Loop networks
    • H04L12/427—Loop networks with decentralised control
    • H04L12/433—Loop networks with decentralised control with asynchronous transmission, e.g. token ring, register insertion
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04J—MULTIPLEX COMMUNICATION
    • H04J14/00—Optical multiplex systems
    • H04J14/02—Wavelength-division multiplex systems
    • H04J14/0227—Operation, administration, maintenance or provisioning [OAMP] of WDM networks, e.g. media access, routing or wavelength allocation
    • H04J14/0254—Optical medium access
    • H04J14/0256—Optical medium access at the optical channel layer
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04J—MULTIPLEX COMMUNICATION
    • H04J14/00—Optical multiplex systems
    • H04J14/02—Wavelength-division multiplex systems
    • H04J14/0227—Operation, administration, maintenance or provisioning [OAMP] of WDM networks, e.g. media access, routing or wavelength allocation
    • H04J14/0254—Optical medium access
    • H04J14/0267—Optical signaling or routing
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04J—MULTIPLEX COMMUNICATION
    • H04J14/00—Optical multiplex systems
    • H04J14/02—Wavelength-division multiplex systems
    • H04J14/0278—WDM optical network architectures
    • H04J14/0283—WDM ring architectures

Definitions

  • Embodiments of the present invention relate to optical devices and arbitration
  • a node can be a processor, memory, a circuit board, a .server, a storage server, an external network connection or any other data processing, storing, or transmitting device. It is often the case that these independently operating nodes need access to the same resources.
  • a resource can be a frequency channel of electromagnetic radiation (i.e., channel) used for transmitting information, or a resource can be a device, such as a broadcast bus. For example, two nodes may need to use a particular channel for transmitting information. Without coordination the two nodes may simultaneously attempt to use the same channel, resulting in lost information sent by both nodes.
  • arbitration In order to coordinate access to shared resources, computer systems often employ a conflict resolution scheme called "arbitration" to prevent two -or more nodes from simultaneously using the same resource.
  • arbitration is used to select and grant one of the nodes access to the resource.
  • Arbitration is critical to computer system performance.
  • arbiters often create latencies that can adversely affect system performance, increase service time, and lower resource utilization.
  • arbiters create unfair sharing of resources and may not scale adequately as the number of nodes and/or resources increase.
  • Figure I shows a table representing "on” and “off states of injectors and detectors in accordance with one or more embodiments of the present in vention.
  • Figure 2 shows an example of a light pulse extracted from a waveguide in accordance with one or more embodiments of the present invention.
  • Figure 3 shows an example of a token-slot arbitration system configured in accordance with one or more embodiments of the present invention.
  • Figures 4A-4B show two example representations of tokens and corresponding slots of a channel in accordance with two or more embodiments of the present invention.
  • Figures 5 A-5C show snapshots of a token-slot arbitration system during an example operation in accordance -with one or more embodiments of the present invention.
  • Figure 6 shows an example of a fair token-slot arbitration system configured in accordance with one or more embodiments of the present invention.
  • Figure 7 shows an example state diagram of three states a requesting node passes through during arbitration in accordance with one or more embodiments of the present invention.
  • Figure 8 sho ws an example state diagram of two states resource passes through during arbitrati on in accordance wi th embodiments of the present invention .
  • Figures A-9D show snapshots of a fair token-slot arbitration system during an example operatio in accordance with one or more embodiments of the present invention.
  • Figure 10 shows an example control-flow diagram summarizing steps taken by a requesting node and a home node in accordance with one or more embodiments of the present invention.
  • Figure 1 1 shows a schematic representation of an example ⁇ ⁇ -core chip with an optical path that passes each core in accordance with one or more embodiments of the present invention.
  • Embodiments are directed to arbitration systems and methods.
  • Embodiments include a continuous, all-optical token-ring arbitration scheme that can be used to assign nodes exclusive access to a shared resource.
  • System embodiments can be implemented with three continuous loop waveguides, each waveguide spanning all of the nodes, minimizing the latency overhead.
  • ai'bitration selects one of the requesting nodes and grants the node access to the resource.
  • ai'bitration system and method embodiments of the present invention satisfy the following requirements.
  • (I) Arbitration is frequent. For example in a cache coherent many-core system, messages can be cache-line size or smaller.
  • Arbitration is fast. Arbitration embodiments do not add unnecessary or excessively to the message transfer time by matching the latency of arbitration with the latency of communications. (3) Arbitration is effective. Arbitration embodiments yield high utilization of resources. End-to-end flow control can further improve bandwidth utilization by avoiding failed transmissions due to buffer overflows. In order to address this issue, embodiments of the present invention aiiow flow control information (i.e., credits) to be included in the arbitration signals. (4) Arbitration is fair.
  • Nodes with equal needs for a resource receive equal access to the resource, and no node is deprived or starved of access to the resource. Bias in arbitration can cause longer queuing delays for packets from an unfairly treated node.
  • Embodiments of the present invention ensure fairness by dynamically adjusting the ai'bitration priorities when the system detects an imbalance in sendees provided to the nodes.
  • the following description is organized into four subsections .
  • the first subsection describes the optical components and introduces optical arbitration.
  • the second subsection provides a general description of optical token-slot arbitration.
  • the third subsection provides a description of optical fair token-slot arbitration system and method embodiments of the present invention.
  • the fourth subsection provides a description of scalabi lity of the arbitration systems and methods .
  • optical refers to devices that operate in both the visible and non- visible portions of the electromagnetic spectrum
  • examples used to describe various system and method embodiments of the present invention are composed of four nodes, these are jus examples and various embodiments of the present invention are in no way restricted to pro viding arbitration for systems having just four nodes.
  • Arbitration system and method embodiments can be scaled up or down to provide arbitration for systems having more, or fewer, than four nodes.
  • System embodiments of the present invention employ optical communication structures such as waveguides and wavelength selective elements ("WSEs").
  • WSEs and. waveguides can be mieroring resonators and ridge waveguides, in other embodiments, the WSEs and waveguides can be photonic crystal resonant cavities and photonic crystal waveguides.
  • a WSE can be configured with a resonance wavelength substantially matching a particular wavelength of light such that by placing the WSE adjacent to and within the evanescent field of light traveling in a waveguide, the WSE evanescently couples the wavelength of light from the waveguide and traps the light for a period of time.
  • the resonance wavelength of a WSE can be electronically switched in and out of resonance with a wavelength of light carried by an adjacent waveguide, enabling the WSE to be operated as a modulator or a detector of light travelling in the adjacent waveguide, or the WSE can be operated to divert the light from the waveguide to another waveguide.
  • a WSE configured to detect light of a particular wavelength carried by an adjacent waveguide is called a "detector,” and a WSE configured to divert light of a particular wavelength into an adjacent waveguide is called an "injector.”
  • Figure 1 shows a table representing "on” and “off * states of injectors and detectors in accordance with one or more embodiments of the present invention. As shown in Figure 1 , injectors are represented by solid rings and detectors are represented by dashed rings. When a WSE is turned “on,” the WSE is resonant with a particular wavelength of light and the light is coupled from an adjacent waveguide into the WSE.
  • the light in the case of the injector, the light can be subsequently injecting into another adjacent waveguide, and in the case of the detector, the light is trapped within the detector and generates an electronic signal that can be used to indicate the light has been extracted from the adjacent waveguide.
  • the WSE when a WSE is turned “off,” the WSE is not .resonant with the wavelength of light and the light passes the WSE unperturbed.
  • the table in Figure I depicts two different "on” states for the injector and the detector iiidentified as "lit” and "not lit.”
  • the plus symbols 102 indicate that the WSE is turned “on” and ready to couple light from a adjacent waveguide, but no light is diverted.
  • the solid circles 104 indicate that the WSE is turned “on” and light is being diverted into the WSE from a adjacent waveguide.
  • the WSE can be configured and operated so that the WSE is "on” when an appropriate voltage is applied to the WSE and when the voltage is no longer applied the WSE is “off.” In other embodiments, the WSE can be configured and operated so. that the WSE is “off” when an appropriate voltage is applied to the WSE and when the voltage is no longer applied the WSE is. “on.”
  • Figure 2 shows an example of a light pulse being extracted from a waveguide 202 in accordance with one or more embodiments of the present invention.
  • the six detectors are configured to have resonance- with the wavelength of a light pulse, represented by directional arrow 204, when turned “on.”
  • two of six detectors 206 and 208 are turned “on.”
  • the pulse 204 travel in one direction along the waveguide 202 and is coupled out of the waveguide 202 by the detector 206, because detector 206 is the first "on” detector the pulse 204 encounters along the waveguide 202. Note that because the detector 206 extracts the pulse 204 from the waveguide 202, the detector 208 remains “on” but is unable to detect the pulse 204.
  • Detectors are configured to extract substantially all of the corresponding light from an adjacent waveguide and generate -a corresponding electronic signal that can be used to confirm that the light has been extracted from the waveguide.
  • the output of a detector is therefore a logical function of the state of all the detectors located closer to the pulse source. This wired-or-like combinatorial operation is performed without any delay other than the time of flight of light in the waveguide.
  • upstream and downstream can be used to describe the relative locations of the detectors and nodes along the waveguide.
  • the node 206 can be referred to as being "upstream” of the node 208, alternatively the node 208 can be referred to as being "downstream” of the node 206.
  • Figure 2 also represents a simple approach to optical arbitration in accordance with one or more embodiments of the present invention.
  • the light pulse 204 can represent a single bit of information, and the waveguide 202 can represent a portion of an arbitration wavegu e.
  • the presence of the light pulse 204 represents the availability of a resource and is referred to as a. ' "token.”
  • a. ' token
  • Each of the six detectors can be electronically coupled to a different node (not shown), Nodes desiring to use the resource turn 'On" thei corresponding detectors.
  • a node that does not want to use the resource leaves its corresponding detector turned "off,"
  • the node electronically coupled to the detector 206 extracts the entire token, gaining exclusive access to the corresponding resource, while the node electronically coupled to the downstrea detector 208 has to wait for the next available token.
  • FIG. 3 shows an example of an arbitration system 300 configured in accordance with one or more embodiments of the present invention.
  • the arbitration system 300 enables each of four nodes PO, PI, P2, and P3 to control access to a particular resource.
  • the arbitration system 300 includes a loop waveguide 302 that passes each of the four nodes.
  • Each node includes an injector located adjacent to and lies within the evanescent fields of light traveling in the waveguide 302.
  • node P0 includes an injector labeled 10
  • node PI includes an injector II, and so on.
  • Each node also includes a set of four detectors labeled DO, Di , D2, and D3. The detectors are also located adjacent to the waveguide 302 and lie within the evanescent fields of light traveling in the waveguide 302.
  • Each injector and each detector is configured to have resonance with a particular wavelength of light when turned “on.”
  • the injectors 10, ⁇ , 12, and 13 are resonant with the wavelengths o, ⁇ [, ⁇ , and ⁇ , respectively, when turned “on”
  • the detectors DO, D l , D2 ⁇ and D3 are also resonant with the wavelengths AQ, X ⁇ , ⁇ , and ⁇ , respectively, when turned "on.”
  • the injectors are operated to place a token (i.e., poise of light) onto the waveguide 302 indicating to the other nodes that the resource controlled ' by the node is available for use.
  • the node P0 turns the injector 10 " n" for a period of time and places a token with the wavelength Q onto the waveguide 302.
  • the tokens can be injected into the waveguide by evanescently coupling light from a power waveguide (not shown) located adjacent to the injector 10.
  • the power waveguide can be configured to loop around the . outside of the arbitration waveguide and be optically coupled to light sources, such as a laser or a light- emitting diode (“LED”), that provide a continuous source of light with the wavelengths 3 ⁇ 46,, 3 ⁇ 4, and A 3 .
  • a light source can be disposed adjacent to the injector 10 and light is evaneseently coupled from the light source into the arbitration waveguide via the injecior ⁇ 0.
  • a node desiring to use resource turns "on" the detector corresponding to the token associated with the resource and waits to extract and detect the token on the waveguide 302.
  • a resource can be located at a particular node, such as an output port, or a resource can be a common or shared resource that can be used by more than one node, such as bus waveguide or a particular frequency of light called a "channel" for transmitting information encoded in optical signals.
  • a node with an available resource is called a "home node,” and any node needing to use a resource is called a “requesting node.”
  • the terms "home node” and "requesting node” are relative terms, and are used to distinguish nodes with an available resource from other nodes needing to use a resource.
  • a first node can be a home node for a second node, and at a later time, the first node can be a requesting node to a third node.
  • the duration of the -token can correspond to a time slot or length of time in which a requesting node gets access to an available resource, or the physical length of the token can correspond to a fixed length or physical slot or portion of a channel.
  • Tokens circulating on the arbitration waveguide are in one-to-one correspondence to the slots.
  • one packet of information can represent one slot.
  • Figure 4A shows a first example of a token and corresponding siot of a channel in accordance with one or more embodiments of the present invention.
  • Figure A includes a local clock signal 402 of a requesting node.
  • Line segment 404 represents a token.
  • the token 404 has the duration of 4 clock cycles, but because the token is a pulse of light traveling in the arbitration waveguide, the token also have a physical length.
  • the length of the token 404 may, in certain embodiments, correspond to the physical length of a slot in the channel used by the requesting node to transmit information to the home node.
  • Rectangle 406 represents a slot in the channel having approximately the same physical length as the corresponding token 404.
  • the requesting node waits for the arrival of the token 404, removes the token 404, and modulates the light of the channel in the corresponding slot 406.
  • the tokens can be spaced about four clock cycles apart.
  • the slot 406 can represent one packet of information. Note thai arrival of the token 404 precedes the slot 406 by some fixed number of clock cycles, such as two or more clock cycles, which are used to set up the data for transmission. Note also thai the token 404 overlaps with two clock cycles of the slot 406. This is permitted provide the token has not also been encoded with information.
  • the channel can be carried to the home node via a separate waveguide.
  • the slot 406 is occupied by modulated light (i.e., data) until the slot gets to the home node, which removes the data, freeing the slot. Meanwhile, the home node continues to inject tokens.
  • Embodiments of the present invention are not limited to the token having the same duration or length as the slot of a channel or the same duration with which a resource can be used.
  • the tokens generated by a home node can have any suitable length.
  • Figure 4B shows a second example of a token and corresponding slot of a channel in accordance with one or more embodiments of the present invention.
  • the token 408 corresponding to the slot 406 is about one clock cycle.
  • the requesting node waits for the arrival of the token 408, removes the token 408, and modulates the light in the corresponding slot 406.
  • FIGS. 5A-5C show snapshots of the arbitration system 300 during an example operation in accordance with one or more embodiments of the present invention.
  • node P0 is the home node that controls access to a resource, such a channel used by nodes PI , P2, and P3 to send information to node P0.
  • a resource such as a channel used by nodes PI , P2, and P3 to send information to node P0.
  • the injector of PO used to place tokens corresponding to the resource is shown and only the detectors of nodes PI , P2, and P3 used to extract the tokens produced by node P0 are shown.
  • node P0 places tokens represented by directional arrows 501 -505 onto the arbitration waveguide 302.
  • the tokens 501 -505 are approximately regularly spaced in time and distance and have approximately the same length.
  • the tokens can be in one-to-one correspondence with slots of the resource.
  • node P2 needs to use the resource controlled by node PO.
  • node P2 turns "on" the detector 508 and begins extracting a toke 510 from the waveguide 302.
  • Node P2 begin using the resource as described above with reference to Figure 4,
  • the token , extracted by the node P2 does no pass node P3.
  • node P3 does not have access to the resource associated with the token 510 extracted by the node P2.
  • the resource is the channel used to transmit information to node P0
  • the token 510 has been extracted by the node P2
  • the corresponding slot in the channel cannot be used by the node P3 to send information to node PO.
  • node P3 wants to send informatio to node PO
  • node P3 has to wait for the next available token, extract the token from the waveguide 302, and modulate the corresponding slot in the channel to encode the information.
  • Token-slot arbitration encodes flow control information by simply emitting a token only when a resource is available.
  • each token signifies a single credit and if a requesting node removes a token, it has also reserved access to the resource at the home node. When access is relinquished, the home node emits a token.
  • upstream nodes i.e., nodes closest to the home node along the waveguide
  • Tokens taken by upstream nodes prevent downstream nodes from gaining access to the needed resource or services provided by the home node.
  • Downstream requesting nodes deprived of a resource by upstream nodes are referred to as "hungry'* nodes. For example, returning to the example shown in Figures 5A-5C, if nodes PI and P2 take the tokens injected into the waveguide by home node P0, requesting node P3 is denied access to the resources and services provided by node P0 and node P3 becomes a "hungry" node.
  • token- slot arbitration described above is inherently unfair because it gives higher priority to upstream nodes.
  • fair token-slot arbitration embodiments of the present invention ensure fairness in that every requesting node is treated nearly equally by detecting hungry nodes and then servicing these nodes once each in some sequence before returning to token-slot arbitration.
  • the essential idea is to switch between token-slot arbitration described above and an alternative that allocates credits to senders in a cyclic manner. In other words, during periods of little demand for a resource, token-slot arbitration describe above is used. However, when higher demand for the resource occurs, requesting nodes may receive less sendee than they need. Fair token-slot arbitration provides fixed quanta of credi ts to the under served requesting nodes, once each cycle, before returning to token-slot arbitration.
  • FIG. 6 shows an example of a fair token-slot arbitration system 600 configured in accordance with one or more embodiments of the present invention.
  • the system 600 includes an arbitration waveguide 602, a hungry waveguide 603, and a broadcast waveguide 604. As shown in Figure 6, the arbitration, hungry, and broadcast waveguides are loop-shaped waveguides that pass each of the nodes.
  • the system 600 enables home node P0 to control access to a resource offered to nodes PI, P2, and P3.
  • the system 600 includes injectors 608-610 and detectors 612-614 electronically coupled to and operated by home node P0.
  • the injectors 608-610 are evanescentiy coupled to the arbitration, hungry, and broadcast waveguides 602-604, respectively, and the detectors 612-614 are also evanescentiy coupled to the arbitration, hungry, and broadcast waveguides 602-604, respectiveiy.
  • the system 600 also includes detectors 616-618 electronically coupled to node PI and evanescentiy coupled to the waveguides 602-604, respectively; detectors 622-624 electronically coupled to node P2 and evanescentiy coupled to the waveguides 602-604, respectively; ' and detectors 626-628 electronically coupled to node P3 and evanescentiy coupled to the waveguides 602-604, respectively.
  • Home node P0 injects tokens into the arbitration waveguide 602 and light into the hungry and broadcast waveguides 603 and 604, the tokens and light traveling in the same direction identified by directional arrow 630.
  • broadcast waveguide 604 includes fractional splitters 632-634 that divert portions of the light carried by the broadcast waveguide 604 to the detectors 620, 624, and 628, respectively.
  • the tokens and light can be injected into the waveguides 602, 603, and 604 by evanescentiy coupling light from separate power waveguides (not shown) located adjacent to the injectors 608-610.
  • the power waveguides can be configured to loop around the outside of the respective waveguides 602-604 and be optically coupled to light sources, such as lasers or LE0s.
  • the- light sources can be disposed adjacent to the. injectors 608-610 and light evanescentiy coupled from the light sources into the waveguides 602-604 via the injectors 608-610, respectively.
  • the WSEs and waveguides can be composed of an elemental semiconductor, such as (“Si”) and germanium (“Ge”), or a ⁇ -V compound semiconductor, where Roman numerals III and V represent elements in the 10a and Va columns of the Periodic Table of the Elements.
  • Compound semiconductors can be composed of column Ilia elements, such as aluminum (“Al”), gallium (“Ga”), and indium (“in”), in combination with column Va elements, such as nitrogen (“N”), phosphorus (“P”), arsenic (“As”), and antimony ( i( Sb”).
  • Compound semiconductors can also be further classified according to the relative quantities of 111 and V elements.
  • binary semiconductor compounds include semiconductors with empirical formulas GaAs, inP, InAs, and GaP; ternary compound semiconductors include semiconductors with empirical formula GaAs v Pj -y . where y ranges from greater than 0 to less than I ; and quaternary compound semiconductors include semiconductors with empirical formula ln A Gaj route x As,,Pi.,,, where both x and y independently range from greater than 0 to less than 1.
  • Other types of suitable compound semiconductors include II- VI materials, where II and VI represent elements in the l ib and Via columns of the periodic table.
  • CdSe, ZnSe, ZnS, and ZnO are empirical formulas of exemplary binary Il-Vi compound semiconductors.
  • the WSEs can be configured as p-i-n junctions and electronically operated to control resonance with particular wavelengths of light traveling i the adjacent waveguides 602-604.
  • an intrinsic WSE can be electronically operated by doping inner and outer regions of the substrate surrounding the WSE with appropriate p-type and n-typc impurities.
  • P-type impurities cart be atoms that introduce vacant electronic energy levels called "holes" to the electronic band gaps of the WSE. These impurities are also called “electron acceptors.”
  • N-type impurities can be atoms that introduce filled electronic energy levels to the electronic band gap of the WSE. These impurities are called “electron donors.”
  • Al, and Ga are p-iype impurities that introduce vacant electronic energy levels near the valence band of Si; and P, As, and Sh are n-type impurities that introduce filled electronic energy levels near the conduction band of Si.
  • column VI impurities substitute for column V sites in the III-V lattice and serve as n-type impurities, and column II impurities substitute for column ill atoms in the III-V lattice to form p- type impurities.
  • Moderate doping of the WSE can have impurity concentrations in excess of about 10 15 impurities/cm 3 while more heavy doping of the core 404 can have impurity concentrations in excess of about 10 i impurities/cm "1 .
  • the waveguides and WSEs of the system 600 can be formed using nanoimprinl lithography, reactive ion etching, focused beam milling or any other well- known method for forming nanoscale optical devices.
  • Figure 7 shows an example state diagram of three states a requesting node passes througli during arbiiration i accordance with one or more embodiments of the present invention. As shown in the example of Figure 7, each requesting node is in one of three possible states: satisfied 701 , hungry 702, and suspended 703. As indicated by directional arrows 706-708, a requesting node cyclically traverses the state diagram from satisfied 701 to hungry 702 to suspended 703.
  • a requesting node is in a satisfied state 701 when the node can extract a token from the arbitration waveguide and use the resource accordingly.
  • the requesting node enters the hungry state 702 when the node must access a resource or other services provided by the home node but is unable to.
  • a satisfied node can enter the hungry state 702 according to a local criterion, such as the node is storing at least one old packet that has been stored at the node longer than some threshold of time ⁇ , or when its input queue length exceeds some queue length threshold, because there may be an upper bound L on the number of packets in the queue of a requesting node thai enters the hungry state.
  • the requesting node enters the suspended state 703 in which the node passes o all tokens.
  • Hungry nodes do not remain indefinitely in the hungry state and their time in the hungry state can be bound.
  • the suspended state enables other hungry downstream nodes to in turn access the tokens. Passing througli the states 701-703 i fair token-slot arbitration is described in greater detail below with regard to a specific example in Figures 9A-9D.
  • Figure 8 shows an example state diagram of the two states a resource passes through during arbitration in accordance with embodiments of the present invention.
  • a resource is controlled by the home node and goes througli alternating plenty and famine states 802 and 804 assigned by the home node.
  • the famine state 802 occurs when the home node receives notification via the hungry waveguide that at least one requesting node is in the hungry state 702.
  • the famine state 804 is communicated to the nodes using the broadcast waveguide where all nodes except for the hungry requesting nodes stop attempting to extract tokens from the arbitration waveguide.
  • Tokens injected into the arbitration waveguide during the famine state are called "famine tokens.”
  • famine tokens During the famine state 804 only the hungry requesting nodes can extract famine tokens while the other nodes let the famine tokens pass unperturbed.
  • each hungry nodes gains access to the resources in sequential order from closest upstream hungry node to the farthest downstream hungry node.
  • the hungry node enters the suspended state. Once all of the hungry nodes have finished using the resource and entered the suspended state 703, the home node receives notification over the hungry waveguide and the resource enters the plenty state 802.
  • Tokens injected into the arbitration waveguide during the plenty state are called "plenty tokens.” Token-slot arbitration resumes and any node can take an available plenty token. Passing through the states 802 and 804 and using the hungry and broadcast waveguides in fair token-slot arbitration is described in greater detail below with regard to a specific example in Figures 9A-9D.
  • FIGS 9A-9D show snapshots of the arbitration system 600 during an example operation in accordance with one or more embodiments of the present invention, hi the following 'description
  • the resource is a channel used by the nodes PI , P2, and P3 to send information to the node P0.
  • node P0 is the home node and nodes P I , P2, and P3 can be requesting nodes.
  • the channel can be carried on a separate waveguide (not shown) that passes each of the nodes PL P2, and P3 and terminates at the node P0.
  • the arbitration systems and methods described below are not limited to providing arbitration of channels, but ca also be used to control access to other resources, such as buses, input and output ports, and oilier shared devices and services provided by a home node.
  • node P0 turns "on" injector 609 in order to inject a continuous stream of light 902 into the hungry waveguide 603 and turns "on" detector 613 i order extract from, and confirm the presence of, the light 902 on die hungry waveguide 603. While light is detected at the detector 613, token-slot arbitration is performed on the arbitration waveguide 602 as described above with reference to Figures 4-5. Node P0 turns "on" detector 612 in order to extract any unextracted tokens from the arbitratio waveguide.
  • node PI is in the process of extracting the token 904, gaining exclusive access to the corresponding slot in the channel for sending information to the node PO.
  • nodes PI , P2, and P3 have tamed "on" detectors 620, 624, and 628, respectively, and wait for a broadcast from node PO on broadcast waveguide 604.
  • node P2 has entered the hungry state because packets destined for node P0 cannot be sent until node P2 can extract tokens from the arbitration waveguide. In other words, packets have been stored at node P2 longer than F or the queue length of node P2 is greate than L,
  • node P2 transitions from the satisfied state 701 to the hungry- state 702 and marks the paekets. stored in- the input queue for the channel. As shown in Figure 9B, node P2 turns "on" the detectors 622-624 and begins extracting the light 902 from the hungry waveguide 603.
  • node P0 when node PO no longe detects light at the detector 613, node P0 places the channel into the famine state 804 and begins broadcast the famine state to the nodes P I . P2, and P3 on the broadcast waveguide.
  • node P0 when node P0 enters the channel into the ' famine state, node P0 turns "on" injector 610 and detector 614 and injects light 906 into the broadcast waveguide 604. Portions of the light 906 are diverted by the detectors 620, 624, and 628 altering nodes PI, P2, and P3 that the channel is the famine state:.
  • node P 1 rums "off detector 618 and node P3 leaves detector 626 tamed “off.”
  • node P2 leaves on detectors 622 and 623 and extracts a token 908 from the arbitration waveguide and continues to extract light 902 from the hungry waveguide.
  • Node P2 extract tokens consecutively until the marked packets are flushed. Any packets thai arrive at node P2 after the hungry state has been declared join the queue, but these packets are not marked and sent during tire famine state.
  • node P2 Once node P2 sends the last of the marked packets, node P2 enters the suspended state 703 shown in Figure 7, in which node P2 turns "off detectors 623 and 622 allowing light 902 to pass on the hungry waveguide 603 and the node P2 passes on all famine tokens, or tokens sent during the famine state.
  • node P0 detects light on the hungry waveguide, node P0 re-enters the plenty state 802.
  • node P2 has turned "off detectors 622 and 623 restoring light 902 on the hungry waveguide 603.
  • node P0 When node P0 detects the light at detector 613, node P0 turns "off injector 610, nodes PI, P2, and P3 subsequently no long detect a broadcast on the broadcast waveguide 604, and the node P2 can enter the satisfied state and token-slot arbitration is again restored. On the next clock cycle node P2 can transition to the hungry state if it needs to.
  • no node can remain in the hungry state indefinitely.
  • the term "current famine” means the epoch of consecutive clock cycles, including the clock cycles at which node A's hunger signal arrives at the home node, over which the home node emits famine tokens.
  • the famine tokens of the current famine may be taken by othe upstream hungry nodes that subsequently enter the himgry state while the node h i s in the hungry state.
  • nodes will take more than L of the tokens during the current famine, because any node that takes L must have flushed its buffer and entered into a suspended state, keeping, newly arrived packets in the queue. After the previously hungry upstream nodes have entered the suspended state, node h will eventually receive' ' enough packets to flush the marked packets fr m its input queue, enter the suspended state, and de-assert hunger. No node returns to the hungry state before ail the nodes that entered the hungry state when the node became hungry enter the suspended state.
  • FIG 10 shows an example control flow diagram summarizing the steps taken by a requesting node and home node m transitioning through the state diagrams shown in Figures 7 and 8 in accordance with one or more embodiments of the present invention
  • the home node injects tokens into the arbitration waveguide.
  • the home injects light into the hungry waveguide.
  • the requesting node identifies the criteria for entering the hungr state. For example, the requesting node may realize that the packet wait time is greater than W or the input queue length is greater than L.
  • the requesting node turns "on" the detector that diverts light from the hungry waveguide, as described above with reference to Figure B.
  • step 1005 the home node detects no light on the hungry waveguide and enters the famine state, in step 1006, the home node broadcast on the broadcast waveguide causing other non-hungry nodes to stop taking tokens from the arbitration waveguide, as described above with reference to Figure 9C.
  • step 1007 the hungry requesting node extracts tokens from the arbitration waveguide.
  • step 1008 the requesting node uses the resource.
  • step 1010 the home node detects light on the hungry waveguide and enters the plenty state.
  • step 1011 the home node stops broadcasting on the broadcast waveguide.
  • the requesting node no longer detects light on the broadcast waveguide indicating ..that the famine state is over and enters the satisfied state where the node can extract tokens from the arbitration waveguide.
  • the requesting node can use the resource when tokens are available.
  • Arbitration system and method embodiments of the present invention are not limited to systems having just four nodes. Arbitration systems and methods can be scaled up or down in order to accommodate any number of nodes. For example, system and method embodiments can be applied to multicore processors. Because chip area is essentially fixed as core count grows with Moore's Law, each core in an iV-core chip has an area proportional to l/N and has a length and width proportional to I N , The arbitration, hun r , broadcast and power waveguides can be configured to follow a path that visits all N cores.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)
  • Optical Communication System (AREA)

Abstract

Selon divers modes de réalisation, la présente invention concerne des systèmes et des procédés d'arbitrage. Dans un mode de réalisation, un système d'arbitrage comporte un guide d'onde d'arbitrage en boucle (602), un guide d'onde démuni en boucle (603), et un guide d'onde de diffusion en boucle (604). Les guides d'onde d'arbitrage, démuni et de diffusion sont optiquement couplés à un noeud principal et à un certain nombre de noeuds demandeurs. Le guide d'onde d'arbitrage transmet des jetons injectés par le noeud principal. Un jeton extrait par un noeud demandeur octroie au noeud un accès à une ressource pendant la durée ou la longueur du jeton. Le guide d'onde démuni transmet de la lumière injectée par le noeud principal. Un noeud demandeur dans un état démuni extrait la lumière du guide d'onde démuni. Le guide d'onde de diffusion transmet de la lumière injectée par le noeud principal de telle manière que la lumière indique à des noeuds demandeurs qui ne sont pas dans l'état démuni d'arrêter d'extraire des jetons du guide d'onde d'arbitrage.
PCT/US2009/067574 2009-12-10 2009-12-10 Systèmes et procédés d'arbitrage par jetons équitable WO2011071499A1 (fr)

Priority Applications (5)

Application Number Priority Date Filing Date Title
EP09852137.0A EP2510448B1 (fr) 2009-12-10 2009-12-10 Systèmes et procédés d'arbitrage par jetons équitable
PCT/US2009/067574 WO2011071499A1 (fr) 2009-12-10 2009-12-10 Systèmes et procédés d'arbitrage par jetons équitable
US13/386,382 US8564867B2 (en) 2009-12-10 2009-12-10 Fair token arbitration systems and methods
CN200980161282.0A CN102483727B (zh) 2009-12-10 2009-12-10 公平令牌仲裁系统和方法
TW099143002A TWI430103B (zh) 2009-12-10 2010-12-09 公平符記仲裁系統與方法

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/US2009/067574 WO2011071499A1 (fr) 2009-12-10 2009-12-10 Systèmes et procédés d'arbitrage par jetons équitable

Publications (1)

Publication Number Publication Date
WO2011071499A1 true WO2011071499A1 (fr) 2011-06-16

Family

ID=44145824

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2009/067574 WO2011071499A1 (fr) 2009-12-10 2009-12-10 Systèmes et procédés d'arbitrage par jetons équitable

Country Status (5)

Country Link
US (1) US8564867B2 (fr)
EP (1) EP2510448B1 (fr)
CN (1) CN102483727B (fr)
TW (1) TWI430103B (fr)
WO (1) WO2011071499A1 (fr)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP6310260B2 (ja) * 2014-01-20 2018-04-11 株式会社荏原製作所 基板処理装置内の複数の処理ユニットを調整するための調整装置、および該調整装置を備えた基板処理装置
CN117538997A (zh) * 2022-08-01 2024-02-09 杭州光智元科技有限公司 光互连件及其制造方法、芯片封装
CN115686864B (zh) * 2022-11-17 2023-09-15 沐曦集成电路(南京)有限公司 一种仲裁系统

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6823140B1 (en) * 1997-03-26 2004-11-23 Sun Microsystems, Inc. Optical computer bus with dynamic bandwidth allocation
US20080183927A1 (en) * 2007-01-31 2008-07-31 Broadcom Corporation, A California Corporation Shared RF bus structure
WO2009055032A2 (fr) 2007-10-23 2009-04-30 Hewlett-Packard Development Company, L.P. Arbitrage réparti rapide entièrement optique dans un dispositif de système informatique

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5751715A (en) * 1996-08-08 1998-05-12 Gadzoox Microsystems, Inc. Accelerator fiber channel hub and protocol
CN101473603B (zh) * 2006-06-26 2014-11-26 三菱电机株式会社 通信节点以及环状通信系统的令牌发行方法、令牌环通信方法

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6823140B1 (en) * 1997-03-26 2004-11-23 Sun Microsystems, Inc. Optical computer bus with dynamic bandwidth allocation
US20080183927A1 (en) * 2007-01-31 2008-07-31 Broadcom Corporation, A California Corporation Shared RF bus structure
WO2009055032A2 (fr) 2007-10-23 2009-04-30 Hewlett-Packard Development Company, L.P. Arbitrage réparti rapide entièrement optique dans un dispositif de système informatique

Non-Patent Citations (7)

* Cited by examiner, † Cited by third party
Title
BEAUSOLEIL ET AL.: "A nanophotonic interconnect for high-performance many-core computation", 5TH IEEE INTERNATIONAL CONFERENCE ON GROUP IV PHOTONICS, 17 September 2008 (2008-09-17)
DANA VANTREASE ET AL.: "Corona: System Implications of Emerging Nanophotonic Technology", PROCEEDINGS OF THE 35TH INTERNATIONAL SYMPOSIUM ON COMPUTER ARCHITECTURE (ISCA-35), June 2008 (2008-06-01), BEIJING, CHINA, XP031281675 *
LINJIE ZHOU ET AL.: "Design and evaluation of an arbitration-free passive op tical crossbar for on-chip interconnection networks", APPL PHYS A (2009), vol. 95, 20 February 2009 (2009-02-20), pages 1111 - 1118, XP019710613 *
LINJIE ZHOU ET AL.: "Design and evaluation of an arbitration-free passive optical crossbar for on-chip interconnection networks", APPLIED PHYSICS A; MATERIALS SCIENCE & PROCESSING, 20 February 2009 (2009-02-20)
MARSAN ET AL.: "All-optical WDM multi-rings with differentiated QoS", IEEE COMMUNICATIONS MAGAZINE, vol. 37, no. 2, 1 February 1999 (1999-02-01), pages 58 - 66, XP000804505, DOI: doi:10.1109/35.747250
R.G. BEAUSOLEIL ET AL.: "A Nanophotonic Interconnect for High-Performance Many-Core Computation", 2008 LEOS NEWSLETTER, 21 March 2008 (2008-03-21), XP031344531 *
See also references of EP2510448A4

Also Published As

Publication number Publication date
EP2510448A4 (fr) 2015-07-29
TWI430103B (zh) 2014-03-11
CN102483727B (zh) 2015-02-11
EP2510448A1 (fr) 2012-10-17
TW201140326A (en) 2011-11-16
CN102483727A (zh) 2012-05-30
US20120120468A1 (en) 2012-05-17
US8564867B2 (en) 2013-10-22
EP2510448B1 (fr) 2017-09-20

Similar Documents

Publication Publication Date Title
US8340517B2 (en) Systems and methods for on-chip data communication
Hamedani et al. QuT: A low-power optical network-on-chip
US20160156999A1 (en) Methods and systems for board level photonic bridges
Koohi et al. All-optical wavelength-routed NoC based on a novel hierarchical topology
US20130148506A1 (en) Bufferless nonblocking networks on chip
US8285140B2 (en) Shared-source-row optical data channel organization for a switched arbitrated on-chip optical network
Chen et al. A hierarchical optical network-on-chip using central-controlled subnet and wavelength assignment
Gu et al. MRONoC: A low latency and energy efficient on chip optical interconnect architecture
EP2510448B1 (fr) Systèmes et procédés d'arbitrage par jetons équitable
US8406623B2 (en) Data channel organization for a switched arbitrated on-chip optical network
Liu et al. Wavelength-reused hierarchical optical network on chip architecture for manycore processors
CN105812063A (zh) 基于统计复用的光片上网络系统及通信方法
US20130266315A1 (en) Systems and methods for implementing optical media access control
Diao et al. Packet rescheduling in WDM star networks with real-time service differentiation
Li et al. A large-scale nesting ring multi-chip architecture for manycore processor systems
Fusella et al. Path setup for hybrid NoC architectures exploiting flooding and standby
Cerutti et al. Fast scheduling based on iterative parallel wavelength matching for a multi-wavelength ring network-on-chip
Bonetto et al. Circuit switching and time-domain optical sub-wavelength switching technologies: Evaluations on the power consumption
Bianco et al. A-posteriori access strategies in all-optical slotted WDM rings
Abadal et al. Graphene-enabled hybrid architectures for multiprocessors: Bridging nanophotonics and nanoscale wireless communication
Chen et al. Wavelength assignment in Optical Network-on-Chip: Design and performance
Kunert et al. Providing efficient support for real-time guarantees in a fibre-optic AWG-based network for embedded systems
Hendry Architectures and design automation for photonic networks on chip
Crespo et al. Silicon photonic networks: Signal loss and power challenges
Kamchevska et al. The Hi-Ring Architecture for Data Center Networks

Legal Events

Date Code Title Description
WWE Wipo information: entry into national phase

Ref document number: 200980161282.0

Country of ref document: CN

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 09852137

Country of ref document: EP

Kind code of ref document: A1

REEP Request for entry into the european phase

Ref document number: 2009852137

Country of ref document: EP

WWE Wipo information: entry into national phase

Ref document number: 2009852137

Country of ref document: EP

WWE Wipo information: entry into national phase

Ref document number: 13386382

Country of ref document: US

NENP Non-entry into the national phase

Ref country code: DE