WO2016070813A1 - Procédé et appareil de pont non transparent pour configurer des réseaux pci-express de grandes dimensions - Google Patents

Procédé et appareil de pont non transparent pour configurer des réseaux pci-express de grandes dimensions Download PDF

Info

Publication number
WO2016070813A1
WO2016070813A1 PCT/CN2015/093802 CN2015093802W WO2016070813A1 WO 2016070813 A1 WO2016070813 A1 WO 2016070813A1 CN 2015093802 W CN2015093802 W CN 2015093802W WO 2016070813 A1 WO2016070813 A1 WO 2016070813A1
Authority
WO
WIPO (PCT)
Prior art keywords
address
transparent bridge
switch
switches
data packet
Prior art date
Application number
PCT/CN2015/093802
Other languages
English (en)
Inventor
Norbert Egi
Robert Lasater
Guangyu Shi
Thomas Boyle
Original Assignee
Huawei Technologies Co., Ltd.
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 Huawei Technologies Co., Ltd. filed Critical Huawei Technologies Co., Ltd.
Priority to CN201580059369.2A priority Critical patent/CN107113210B/zh
Priority to EP15856318.9A priority patent/EP3172870B1/fr
Publication of WO2016070813A1 publication Critical patent/WO2016070813A1/fr

Links

Images

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]
    • H04L12/40—Bus networks
    • H04L12/40143—Bus networks involving priority mechanisms
    • H04L12/4015—Bus networks involving priority mechanisms by scheduling the transmission of messages at the communication node

Definitions

  • the present disclosure relates generally to the field of communication and more specifically to the field of network communication.
  • PCI Peripheral Component Interconnect
  • PCIe PCI-Express
  • PCIe technology was originally designed as an onboard interconnection technology.
  • Conventional PCIe environments or networks typically have only a single PCIe switch and a relatively few number of nodes (e.g., CPUs and I/O devices) .
  • nodes e.g., CPUs and I/O devices
  • some conventional PCIe approaches attempt to be a high throughput, low-latency, packet based and switched interconnection technology.
  • the types and complexity of topologies that can be implemented with conventional PCIe approaches are essentially limited (e.g., a small basic tree topology with relatively few levels and end-points, etc) .
  • Figure 1 is a block diagram of one example of a conventional attempt at a basic tree topology.
  • switch 115 and switch 127 with relatively numerous multiple intervening hops (e.g., to switches 121, 122, 124, etc. ) that communications must traverse.
  • end-point 181 and end-point 191 again with numerous multiple intervening hops (e.g., to switches 129, 128, 121, 115, 131, 133, etc. ) that communications must traverse.
  • paths between switches and end-points become too long (e.g., in terms of hops, etc. ) and components (e.g., links close to the root, etc. ) become overloaded.
  • a PCIe network for communicating information includes a non-transparent bridge circuit, a plurality of switches and a plurality of end-point devices.
  • the non-transparent bridge circuit comprises: a first network connection, a second network connection, a processing circuit and a memory.
  • the first network connection is connected to a first switch in a first tree branch of the PCIe network and the second network connection is connected to a second switch in a second tree branch of the PCIe network.
  • the non-transparent bridge circuit is assigned a first range of addresses with respect to the first switch and a second range of addresses with respect to the second switch.
  • the processing circuit is configured to: receive a first data packet from the first switch, the first data packet includes a first payload portion and a first destination address that is within the first range of addresses assigned to the non-transparent bridge; translate the first destination address to a first translated destination address, wherein the first translated destination address belongs to an endpoint device in the second tree branch and connected to the second switch; generate a first modified data packet including the first payload portion and the first translated destination address; and forward the first modified data packet to the second switch for delivery to the endpoint device in the second tree branch. Forwarding the modified data packet to the second switch can bypass an intervening third switch in the second branch and remains PCIe compliant.
  • the processor can also be configured to: receive a second data packet from the second switch, the second data packet includes a second payload portion and a second destination address that is within the second range of addresses assigned to the non-transparent bridge; translate the second destination address to a second translated destination address, wherein the second translated destination address belongs to another endpoint device in the first tree branch and connected to the first switch; generate a modified data packet including the second payload portion and the second translated destination address; and forward the second modified data packet to the first switch for delivery to the endpoint device in the first tree branch.
  • the memory stores information for the processing circuit.
  • the memory can store an address translation table that maps the first destination address to the first translated destination address. A single entry in the address translation table can be utilized to store the first destination address contained in the first data packet.
  • the non- transparent bridge circuit can have a variety of implementations (e.g., a chip, an ASIC, a CPU, etc, ) in various configurations (e.g., separate entity, incorporated in one of the plurality of switches, etc. ) .
  • an apparatus participates in a communication process or method.
  • the method is performed in a PCIe compliant network.
  • Data packets are transmitted via a non-transparent bridge circuit in a PCIe compliant network to a plurality of switches and endpoints coupled to the non-transparent bridge circuit.
  • a data packet from a first one of the plurality of switches in the PCIe compliant network is received in a processing component of the non-transparent bridge circuit.
  • the data packet includes a first address that is within one of a pair of address ranges associated with the non-transparent bridge circuit.
  • the pair of address ranges assigned to the non-transparent bridge circuit can be assigned by a PCIe root complex controller (which can consider the pair of address ranges to be end-point address ranges) .
  • the first address is replaced with a second address to form a modified data packet
  • the second address is associated with a second one of the switches included in the plurality of switches in the PCIe compliant network.
  • the modified data packet is forwarded from the processing component to the second switch via a first data path between the first switch and the second switch.
  • the forwarding can bypass an intervening third switch in a second data path connected to the first one of the plurality of switches and the second one of the plurality of switches.
  • the first data path and second data path enable multi-path routing in the PCIe compliant network
  • the first data path and second data path enable multi-path routing in the PCIe compliant network.
  • the first data path can be at least one hop less than the second data path in the PCIe compliant network.
  • Mapping can prevent data packets from being transmitted through a loop transmission path formed at least in part by the first data path and the second data path.
  • the first address is mapped to the second address, and the mapping is utilized to translate the first address with the second address.
  • the modified data packet is generated by the replacing of the first address in the data packet with the second address.
  • the addresses and address ranges can have a variety of configurations.
  • the size of the one of the pair of address ranges can be equal to or larger than the difference between a beginning address and an end address allocated to a plurality of resources coupled to the non-transparent bridge.
  • the size of the one of the pair of address ranges can be equivalent to the sum of addresses allocated to a plurality of resources coupled to the non-transparent bridge.
  • the second switch is included in the plurality of resources.
  • the length of one of the pair of address ranges can be equivalent to the length of addresses assigned to resources on one communication side of the non-transparent bridge and the length of the of another one of the pair of address ranges can be equivalent to the length of addresses assigned to resources on an opposite communication side of the non-transparent bridge.
  • a PCIe compliant network system communicates information though components configured in a topology that includes multiple different paths between two switches.
  • the PCIe compliant network system comprises: a plurality of switch devices coupled together to form a PCIe compliant tree topology with a plurality of branches; a plurality of end point devices that are coupled to the ends of the plurality of branches; and a non-transparent bridge exposed to a first one of the plurality of switches by a first range of addresses assigned to the non-transparent bridge and exposed to a second one of the plurality of switches by a second range of addresses assigned to the non-transparent bridge.
  • the non-transparent bridge communicates information in the PCIe compliant network between the first one of the plurality of switches and second one of the plurality of switches by translating a destination address within the first range of addresses assigned to the non-transparent bridge to a translated destination address associated with a resource coupled to the second one of the plurality of switches and translating a destination addresses within the second range of addresses assigned to the non-transparent bridge to a translated destination address associated with a resource coupled to the first one of the plurality of switches.
  • the non-transparent bridge circuit comprises: a first network connection, a second network connection, a processor and a memory.
  • the first network connection is connected to the first one of the plurality of switches in a first one of the plurality of branches of the PCIe compliant network and the second network connection is connected to a second one of the plurality of switches in a second one of the plurality of branches of the PCIe compliant network.
  • the processing circuit is configured to: receive a data packet from the first one of the plurality of switches; generate a modified data packet based on results of the translating of the destination address within the first range of addresses, ; and forward the modified data packet to the second switch for delivery to the resource.
  • the data packet includes a first payload portion and the destination address that is within the first range of addresses assigned to the non-transparent bridge.
  • the modified data packet includes the first payload portion and the translated destination address associated with the resource coupled to the second one of the plurality of switches.
  • the resource can include an end-point device
  • the first range of addresses and the second range of addresses can be assigned by a master root complex circuit that coordinates address assignment to the plurality of switches, the plurality of end points and the non-transparent bridge.
  • the first range of addresses and the second range of addresses can be a pair of address ranges that enable the non-transparent bridge to simulate a pair of end points.
  • the multiple different paths can include transmission connections between the plurality of switches, the plurality of end point devices, and the non-transparent bridge to form a high-dimensional topology that is PCIe compliant.
  • An alternative path in the multiple different paths other than through the non-transparent bridge can be established between the first one of the plurality of switches and the second one of the plurality of switches.
  • the PCIe compliant network system can include a first PCIe compliant network domain comprised of up to 256 nodes that include the first switch and a second PCIe compliant network domain comprised of up to 256 nodes that include the second switch, and the non-transparent bridge can forward the payload portion across a boundary of the first PCIe compliant network domain and the second PCIe compliant network domain in accordance with the translation of the first address to the second address.
  • Figure 1 is a block diagram of one example of a conventional attempt at a basic tree topology.
  • Figure 2 is a block diagram of an exemplary network in accordance one implementation.
  • Figure 3 is a block diagram of another exemplary network.
  • Figure 4 is block diagram of an yet another exemplary network.
  • Figure 5A is block diagram of an exemplary network topology representation presented in a manner to illustrate characteristics similar to a simple fat-tree approach.
  • Figure 5B is block diagram of an exemplary network topology representation presented in another manner to illustrate characteristics similar to a basic tree approach.
  • Figure 6 is a graphical representation of an exemplary address allocation and mapping.
  • Figure 7 is a block diagram of an exemplary apparatus in accordance with one embodiment.
  • Figure 8 is a block diagram of exemplary operations of an apparatus.
  • Figure 9 is a block diagram of an exemplary communication process in accordance with one embodiment.
  • Figure 10 is a flow chart of an exemplary method for transmitting data packets through a bridge in a PCIe compliant transmission system in accordance with one implementation.
  • Figure 11 is flow chart of an exemplary method in accordance with one embodiment.
  • Figure 12 is a block diagram of yet another exemplary network in accordance one implementation.
  • the descriptions presented herein include approaches that can facilitate utilization of an efficient protocol (e.g., PCIe, etc. ) while enabling implementation of various characteristics and features (e.g., characteristics and features similar to a fat-tree topology, CLOS topology, 2D and 3D topologies, etc. ) that would otherwise not be compatible with the protocol.
  • Implementation of alternative paths can be enabled and utilized with a protocol (e.g., PCI, PCIe, etc. ) that would otherwise not be compatible with the use of alternative paths.
  • a protocol e.g., PCI, PCIe, etc.
  • the enablement of both simple interconnections e.g., similar to a basic tree topology, etc.
  • alternative path interconnections e.g., similar to a fat tree topology, etc.
  • presented systems and methods facilitate utilization of a non-transparent bridge circuit configured as an end-point with respect to communications from at least one device while facilitating transmission of the communications on to at least one other device. Additional explanation of various aspects of the new approaches is set forth in later portions of the detailed description.
  • FIG. 2 is a block diagram of an exemplary network 300A in accordance with one embodiment.
  • Network 300A includes root complex (RC) 310, a plurality of switches (e.g., 315, 321, 322, 323, 324, 325, 327, 328, 329, 331, 332, and 333) , a plurality of end-point devices (e.g., 381, 382, 383, 384, 385, 387, 388, 389, 391, 392, 393, 394, 395, 398, and 399) and non-transparent bridge circuits 351 and 352.
  • Network 300A is compatible with various tree type topology protocols or requirements (e.g., requirements associated with PCI, PCIe, etc. ) .
  • Non-transparent bridge circuits e.g., 351, 352, etc.
  • a new non-transparent bridge circuit can be implemented in a variety of ways.
  • a new configuration approach can be built into a switch (e.g., similar to 315, 321, etc. ) or can be a standalone device with separate connections to another switch (e.g., 351, 352, etc. ) .
  • the non-transparent bridge circuit can be implemented similar to a bridge configuration (e.g., a bridge circuit, a Non-Transparent Bridge (NTB) , etc. ) .
  • a non-transparent bridge circuit e.g., 351, 352, etc.
  • the topology of network 300A includes some similarities with a basic tree topology.
  • network 300A has a single root complex 115 of a tree with multiple branches. Each branch can span multiple hops and at the end of each branch there are leaves.
  • a hop can include multiple communication connections between nodes (e.g., communication connections between root complex, switches, end-point devices, etc. ) .
  • the leaves can include end-point devices (e.g., input/output (I/O) devices, CPUs, accelerators, etc. ) .
  • I/O input/output
  • the root of the network 300A topology is in charge of control and management of the network (hence called Root Complex (RC) ) .
  • RC Root Complex
  • a root complex is considered a master root complex circuit or PCIe root complex controller that coordinates address assignment to the plurality of switches, the plurality of end points and the non-transparent bridge.
  • the PCIe root complex controller can consider a pair of address ranges assigned to a non-transparent bridge to be end-point address ranges.
  • topology of network 300A may have some similarities to a basic tree topology, there are also significant differences such as alternative paths.
  • utilization of a new approach when building a network or system starts with a configuration similar to a basic tree topology and adds extra or alternative connections between switches.
  • a non-transparent bridge circuit can be utilized to facilitate implementation of the alternative multiple paths.
  • network 300A there is a default or basic path formed by switches 321, 322, 324 and 327.
  • switch 321, non-transparent bridge circuit 351 and switch 327 There is an alternative path formed by switch 321, non-transparent bridge circuit 351 and switch 327.
  • a communication from switch 321 to switch 327 (and vise versa) via the alternative path that includes non-transparent bridge circuit 351 is shorter or one less hop than the default path that includes switches 322 and 324.
  • a non-transparent bridge circuit can cross branches.
  • network 300A there is a path from switch 333 to switch 328 formed by switches 331, 315, 321, and 328.
  • a communication from switch 333 to switch 328 (and vice versa) via the alternative path that includes non-transparent bridge circuit 352 is shorter or two less hops than the default path that includes switches 331, 315 and 321.
  • FIG. 3 is a block diagram of an exemplary network 300B in accordance with one embodiment.
  • Network 300B is similar to network 300A except non-transparent bridge 351 is coupled to end point device 389 instead of switch 327.
  • Alternative paths or connections formed by the a new non-transparent bridge circuit are a unique approach that can facilitate characteristics with complexity and capabilities well beyond limitations of conventional approaches.
  • An exemplary implementation in accordance with the new approach can facilitate implementation of characteristics similar to different topologies while still being compatible with a particular protocol that is otherwise traditionally not compatible with the different topology characteristics. For example, facilitating implementation of characteristics (e.g., simple direct connections, etc. ) similar to a basic tree topology and also characteristics (e.g., alternative connections, etc. ) similar to another topology (e.g., fat-tree, CLOS, etc. ) , while still being compatible with a protocol (e.g., PCI, PCIe, etc. ) that is otherwise traditionally not compatible with the different topology characteristics.
  • characteristics e.g., simple direct connections, etc.
  • another topology e.g., fat-tree, CLOS, etc.
  • a protocol e.g., PCI, PCIe, etc.
  • alternative paths that are otherwise not compatible with a PCIe protocol are implemented while communications remain compatible with PCIe protocol requirements (e.g., no looping communications, etc. ) While the configuration of the components and connections may in a sense form a "loop" and a communication packet can reach the components that form the "loop” , the communication packets are prevented from circulating around the loop in a manner that violates the PCIe protocol.
  • a side of a non-transparent bridge circuit or non-transparent bridge circuit is not necessarily utilized to indicate a physical side but rather a side with respect to communication flow.
  • communication interface components e.g., ports, input/output components, etc.
  • communication interface components that are on different communication sides of the non-transparent bridge circuit can be located on the same physical side of a non-transparent bridge circuit component (e.g. ports located on the same physical side of a non-transparent bridge circuit but communicate with different other upstream and downstream devices, etc. ) .
  • upstream and downstream are terms that can be utilized to indicate a relative communication direction or relationship.
  • the direction or relationship can be between the non-transparent bridge circuit and a particular centralized component (e.g., upstream can be associated with a direction towards a root complex, downstream can be associated with a direction away from a root complex, etc. ) .
  • upstream is considered a side or direction from which control and/or configuration communications are received by the non-transparent bridge circuit and downstream is considered a side or direction from which control and/or configuration communications are forwarded or transmitted by the non-transparent bridge circuit.
  • FIG. 4 is a block diagram of an exemplary network 400 topology presented in a manner to illustrate similarities with a simple fat-tree topology approach.
  • Network 400 topology includes root complex 410, non-transparent bridge circuit 470, and switches 420, 430, 440 and 450.
  • the switches e.g., 420, 430, etc.
  • the switches are PCIe compatible switches in accordance with the new approach allowing a simple fat-tree topology to be built using the PCIe compatible switches.
  • the network 400 topology includes connections similar to "default" or basic PCIe path connections, in a sense forming a path similar to a basic tree (with a single 3 hop branch in a path including switches 420, 430, 440 and 450) .
  • the network 400 topology includes an alternative path connection that includes a non-transparent bridge circuit 470 configured in a way that allows communications to happen over this alternative link similar to a fat tree topology approach without breaking the rules or requirements of PCIe (e.g., regarding looping, etc. ) .
  • Network 400 can include various end-point devices or resources coupled to the switches.
  • CPU 405 is a resource that is optionally coupled to switch 420 for performing various processing operations and results of the processing operations can be forwarded to switch 450 via non-transparent bridge 470.
  • a key to enabling such alternative links or paths that facilitates complex and efficient topology characteristics lies in the way the non-transparent bridge circuit is enumerated (e.g., identified, addressed , etc. ) and configured.
  • a non-transparent bridge circuit can be configured as two end-points and the address translation capability of the non-transparent bridge circuit component can be used to interconnect the end-points and to forward packets between other devices.
  • One of the pair of end-point indications or addresses associated with the non-transparent bridge circuit appears as an end-point to at least one device on one side (e.g., upstream, downstream, etc. ) of the non-transparent bridge circuit.
  • the non-transparent bridge circuit can translate the end-point indication or address into an indication or address associated with at least one other device on another side of the non-transparent bridge circuit. In one exemplary implementation, this allows a non-transparent bridge circuit to appear as an end-point to a device on one side of the non-transparent bridge circuit while still being able to communicate with a device on another side.
  • FIGS 5A and 5B are block diagrams of exemplary network 500 topology representations in accordance with one embodiment.
  • Network 500 includes CPU 505, root complex 510, non-transparent bridge circuit 570, and switches 520, 530, 540 and 550.
  • Non-transparent bridge circuit 570 is configured with two end-point indicators or addresses J and K.
  • Non-transparent bridge circuit 570 is considered having two sides (e.g., upstream, downstream, etc. ) and each side is configured or associated with one of the corresponding end-point indications or addresses.
  • one side of the non-transparent bridge circuit 570 has a port that is communicatively coupled to switch 520 and is configured as end-point address J (with respect to communications from switch 520) and the other communication side of the non-transparent bridge circuit is configured as end-point address K (with respect to communications from switch 550) .
  • the non-transparent bridge circuit 570 can facilitate implementation of the system or network with characteristics of multiple different topologies.
  • Figures 5A and 5B present the exemplary network 500 in two different manners to illustrate the similarities with characteristics of two different topologies.
  • Figure 5A is block diagram of an exemplary network 500 topology presented in a manner to illustrate characteristics similar to a simple fat-tree approach.
  • network 500 is similar to network 400.
  • the topology can logically be represented in Figure 5A, with the two ends of the non-transparent bridge circuit having characteristics similar to a fat-tree topology facilitating efficient alternative communication paths (e.g., shorter, less hops, etc. than other paths through switches 530 and 540) .
  • Figure 5B is a block diagram of an exemplary network 500 topology presented in a manner to illustrate characteristics similar to a basic tree approach.
  • the topology can logically be represented in Figure 5B, with the two ends of the non-transparent bridge circuit 570 having characteristics similar to end-points (e.g., 570A and 570B) in a basic tree topology and fully complying with PCI-Express standards.
  • address translation is configured to enable data packets to pass through the non-transparent bridge circuit to and from other components (e.g., switch (es) , CPU (s) , I/O device (s) , etc. ) the addresses assigned to end-points (e.g., J and K, etc. ) as well as the.
  • addresses allocated to J have to correspond (e.g., be equivalent in length, etc.
  • the address allocated to K has to correspond to the address (es) of the resource (s) being accessed through the non-transparent bridge circuit (e.g., that of the switches, end-point devices I/O devices or CPUs attached to switch 520) . Additional explanation of address configurations and translations is set forth in later portions of the detailed description.
  • FIG. 6 is a graphical representation of exemplary address allocation and mapping in accordance with one embodiment.
  • the address allocation and mapping can enable an additional alternative path connection that is compatible with a PCIe protocol.
  • network 700 is similar to network 500.
  • Switch 750 is communicatively coupled through a path (not shown) via switches 730 and 740 and also through an alternative path via non-transparent bridge circuit 770.
  • There are 3 resources e.g., switches, end-point devices, I/O devices, CPUs, etc. ) communicatively coupled to switch 750 and 4 resources communicatively coupled to switch 720.
  • the 3 resources communicatively coupled to switch 750 are enumerated with indications or addresses (e.g., Y0, Y1 and Y2) and 4 resources connected to switch 720 are enumerated with indications or addresses (e.g., X0, X1, X2, and X3) .
  • the resources can be other devices in a network (e.g., similar to switch 327, end-point devices 388 and 389, etc. ) .
  • the new approach enables the resources to communicate or access one another via a non-transparent bridge circuit 770 on an alternative path that utilizes the address translations.
  • Various mapping approaches can be used to map addresses allocated to the non-transparent bridge circuit to the addresses of the resources on the other side of the non-transparent bridge circuit.
  • An address range is allocated to the non-transparent bridge circuit (side K in the example) and the size of the range is equal to or larger than the difference between the beginning of the first address and the end of the last address allocated to the resources on the other side of the non-transparent bridge circuit.
  • An address range (e.g., address range K) allocated to the non-transparent bridge circuit is mapped to an address range (e.g., marked as K’ ) associated with switch 720 that includes addresses for X0, X1, X2 and X3.
  • K e.g., K0, K1, K2 and K3
  • An Address Translation table can be used to map the respective address ranges.
  • Another mapping approach is to use an Address Translation table to map the address range of the non-transparent bridge circuit (e.g., address range J) to the address ranges of the resources next to each other in the mapped address region. That is, the length or size of the address range allocated on one side of the non-transparent bridge circuit (e.g., address range J in this example) is equivalent to the sum of the addresses (e.g., Y0, Y1, and Y2) allocated to the resources on the other side of the non-transparent bridge circuit to be accessed.
  • an Address Translation table to map the address range of the non-transparent bridge circuit (e.g., address range J) to the address ranges of the resources next to each other in the mapped address region. That is, the length or size of the address range allocated on one side of the non-transparent bridge circuit (e.g., address range J in this example) is equivalent to the sum of the addresses (e.g., Y0, Y1, and Y2) allocated to the resources on the other side of the non
  • an entry in the Address Translation table is used to map each resources’ address (e.g., Y0, Y1, and Y2) next to each other in the address space allocated to the non-transparent bridge circuit's other side (e.g., J) .
  • An address range (e.g., address range J) allocated to the non-transparent bridge circuit is mapped to an address range (e.g., marked as J’ ) associated with switch 750 that includes addresses for Y0, Y1 and Y2.
  • portions or sub-ranges in J e.g., J0, J1, and J2 that correspond to Y0, Y1, and Y2 are shown.
  • the mapping ensures that resources can be accessed through the non-transparent bridge circuit by addressing communications to the mapped address of the non-transparent bridge circuit communication interface component (e.g., port, input/output component, etc. ) instead of the original address of the resource which would lead packets to go over the default longer path.
  • the non-transparent bridge circuit communication interface component e.g., port, input/output component, etc.
  • resource X0 can be a CPU (e.g., similar to CPU 405, CPU 505, etc. ) for performing various processing operations and results of the processing operations can be forwarded to switch 750 via non-transparent bridge 770.
  • the CPU X0 can forward a data packet with an address J0 to Non-transparent bridge 770 which translates the address J0 into Y0 and forwards the data packet to switch 750.
  • the allocation and mapping of addresses utilized by the non-transparent bridge circuit can be controlled by the Root Complex of the network.
  • the translation between the addresses can be performed by the non-transparent bridge circuit.
  • the selection and programming of one of the alternative paths in a multi-path approach can be performed by the drivers.
  • Non-transparent bridge 800 can implement a non-transparent bridge configuration process and communication process.
  • Non-transparent bridge 800 includes communication interface components 811, 812 and 813, interconnection component 835, and controller 830 which includes processing component 831 and memory 832.
  • controller 830 is considered a non-transparent bridge circuit.
  • Communication interface components 811, 812, and 813 receive and transmit communications and can be implemented in various ways (e.g. communication ports, input/output terminals, etc. ) .
  • Interconnection component 835 selectively couples the communication interface components 811, 812, and 813 to one another. Controller 830 controls the interconnection component 835.
  • memory 832 stores instructions and data (e.g., address translation tables, etc. ) for performing various operations (e.g., including non-transparent bridge circuit configuration operations, interconnection control operations, etc. ) , and processing component 831 implements the instructions.
  • processing component 831 e.g., a central processing unit (CPU) , an application specific integrated circuit (ASIC) , a processing circuit, etc.
  • memory 832 e.g., a random access memory, a flash memory, a cache, etc.
  • communications interface components 811, 812, and 813 e.g., a port, a terminal, etc.
  • At least one of the communication interface components (811, 812 and 813) is operable to receive communications; controller 830 is operable to translate between addresses associated with the communications, wherein translation includes translating between at least one address that at least one device considers an end-point address and an address of at least one other device; and at least one of the communication interface components (e.g., 811, 812, 813) is operable to forward communications in accordance with the translation.
  • Non-transparent bridge 800 is coupled to various other components or resources including switches 801, 802, 803, 804, and 805. There are two communication paths between switch 801 and switch 804. One communication path includes switches 801, 802, 803, and 804. There is another communication path that includes switch 801, non-transparent bridge 800 and switch 804. Non-transparent bridge 800 is associated with or assigned at least one pair of end-point addresses (e.g., address M and address N) . Switches 801, 802, 803, 804, and 805 are associated with or assigned addresses T, V, Q, R and W respectively (shown in parenthesis in the Figure 7) . Data packet 841 is sent from switch (T) 801 to Non-transparent bridge 800 ) and data packet 842 is sent from non-transparent bridge 800 to switch (R) 804.
  • T switch
  • R switch
  • FIG. 8 is a block diagram of exemplary operations of non-transparent bridge 800 in accordance with one implementation.
  • Memory 832 operations include storing an address translation table 850.
  • Address translation table 850 includes end-point address column 851 and resource address column 852.
  • End-point address column 851 includes end-point address M and end-point address N and resource address column 852 correspondingly includes address R and address T.
  • Data packet 841 is received from communication interface component 811.
  • the processor 831 directs removal of address M from the packet 841 and uses the address translation table to look up a mapping of address M to address R, which is attached to payload G to form data packet 842.
  • Data packet 842 is forwarded from communication interface 813 to switch 804 in accordance with address R.
  • Data packet 843 is received from communication interface component 813.
  • the processor 831 directs removal of address N from the packet 843 and uses the address translation table to look up a mapping of address N to Address T, which is attached to payload Z to form data packet 844.
  • Data packet 844 is forwarded from communication interface 811 to switch 801 in accordance with address T.
  • a non-transparent bridge circuit (e.g., similar to non-transparent bridge circuit 800, etc. ) can be a bridge circuit included in a PCIe network.
  • the PCIe network can include a plurality of switches coupled together to transmit packets between end-point devices coupled to the switches and the bridge circuit.
  • the bridge circuit (e.g., similar to 351, 352, etc. ) can include a processor that uses an input port to accept a data packet originating from a source end-point device from a first switch (e.g., similar to 389, 327 , etc. ) .
  • the processor examines an address contained in the data packet and, based on the address, directly forwards the data packet to a second switch (e.g., similar to 321, etc. ) , bypassing an intervening switch (e.g., similar to 324, 322, etc. ) which is part of a tree topology.
  • the processor can use an output port to transmit the data packet to the second switch which transmits the data packet to a destination end-point device. Bypassing the intervening switch and forwarding the data packet to the second switch is PCIe compliant.
  • the processor can establish a topology configuration corresponding to two end-points (e.g. similar to 570A and 570B in Figure 5, etc. ) and a data path is created between the two end-points.
  • the data path is used to directly forward the data packet to the second switch.
  • the data path can include a multi-path connection in the tree topology and is PCIe compliant.
  • the bridge circuit also includes a memory that stores information and instructions for the processor.
  • the memory can store an address translation table that maps the address contained in the data packet to an address associated with the second switch, wherein the address in the data packet is one of a pair of a PCIe compliant end-point addresses (e.g., J and K in Figure 6, etc. ) associated with the bridge and the data packet is transmitted through the data path by addressing the address associated with the second switch stored in the address translation table.
  • a single entry in the address translation table can be utilized to store the address contained in the data packet.
  • a non-transparent bridge circuit can enable a multi-path PCIe transmission system in accordance with a new approach described herein.
  • the non-transparent bridge circuit can include a processor that uses an input port to accept a data packet (e.g., originating from a source end-point device from a first switch, etc. ) .
  • the processor examines an address contained in the data packet and, based on the address, forwards the data packet to a second switch, wherein the second switch is capable of being coupled to a third switch which can be used to also route the data packet from the first switch to the second switch.
  • the processor uses an output port to transmit the data packet to the second switch which transmits the data packet to a destination end-point device.
  • the first address can fall within an address range allocated to the non-transparent bridge circuit.
  • the size of the address range is equal to or larger than the difference between a beginning address and an end address allocated to a plurality of resources (e.g., switches, end-point devices, etc. ) couple to the non-transparent bridge circuit, the second switch included in the plurality of resources.
  • the size of the address range is equivalent to the sum of addresses allocated to a plurality of resources couple to the non-transparent bridge circuit, the second switch included in the plurality of resources.
  • the non-transparent bridge circuit can also include a memory that stores information and instructions for the processor.
  • the processor utilizes an address translation table stored in the memory and a single entry in the address translation map is used for the first address.
  • the first address falls within an address range allocated to the non-transparent bridge circuit and the size of the address range is equivalent to the sum of addresses allocated to a plurality of resources couple to the non-transparent bridge circuit, the second switch included in the plurality of resources.
  • the length of the first address is equivalent to the length of addresses assigned to resources on one communication side of the non-transparent bridge circuit and the length of the third address is equivalent to the length of addresses assigned to resources on an opposite communication side of the non-transparent bridge circuit.
  • the non-transparent bridge circuit can be a separate entity or the non-transparent bridge circuit can be incorporated in a switch.
  • a non-transparent bridge circuit participates in a communication process.
  • Figure 9 is a block diagram of communication process 900 in accordance with one embodiment. The communication is between two devices. The first device is on one communication side of the non-transparent bridge circuit and the second device is on another communication side of the non-transparent bridge circuit (e.g., information is received from an upstream device and forwarded to a downstream device, or vice versa, etc. ) .
  • communication process 900 is performed in a PCIe compliant environment. Data packets are transmitted via a non-transparent bridge circuit in a PCIe compliant network including a plurality of switches and endpoints coupled to the non-transparent bridge circuit
  • a data packet from a first one of the plurality of switches in the PCIe compliant network is received in a processing component of the non-transparent bridge circuit, the data packet includes a first address that is first address that is within one of a pair of address ranges associated with the non-transparent bridge circuit.
  • the a pair of address ranges associated with the non-transparent bridge circuit can be assigned by a PCIe root complex controller.
  • the PCIe root complex controller can consider the pair of address ranges to be end-point address ranges
  • the second address is associated with a second one of the switches included in the plurality of switches in the PCIe compliant network.
  • the first address can be mapped to the second address and the mapping can be utilized to translate the first address to the second address.
  • the modified data packet is generated by the replacing of the first address in the data packet with the second address.
  • the mapping can prevent data packets from being transmitted through a loop transmission path formed at least in part by the first data path and the second data path.
  • the addresses and address ranges can have a variety of configurations.
  • the size of the one of the pair of address ranges can be equal to or larger than the difference between a beginning address and an end address allocated to a plurality of resources coupled to the non-transparent bridge.
  • the size of the one of the pair of address ranges can be equivalent to the sum of addresses allocated to a plurality of resources coupled to the non-transparent bridge.
  • the second switch is included in the plurality of resources.
  • the length of one of the pair of address ranges can be equivalent to the length of addresses assigned to resources on one communication side of the non-transparent bridge and the length of the of another one of the pair of address ranges can be equivalent to the length of addresses assigned to resources on an opposite communication side of the non-transparent bridge.
  • the second data packet is forwarded from the processing component to the second switch via a first data path between the first switch and the second switch, bypassing an intervening third switch in a second data path connected to the first one of the plurality of switches and the second one of the plurality of switches.
  • the first data path and second data path enable multi-path routing in the PCIe compliant network.
  • the first data path can be at least one hop less than the second data path in the PCIe compliant network.
  • Figure 11 is a flow chart of exemplary apparatus configuration process or method 1100 in accordance with one embodiment of the present invention.
  • the apparatus is compatible with Peripheral Component Interconnect Express (PCIe) protocol requirements.
  • PCIe Peripheral Component Interconnect Express
  • the apparatus can emulate at least some non basic tree topology characteristics (e.g., characteristics similar to a fat-tree topology, CLOS topology, 2D topology, 3D topology, etc. ) .
  • PCIe Peripheral Component Interconnect Express
  • the apparatus can emulate at least some non basic tree topology characteristics (e.g., characteristics similar to a fat-tree topology, CLOS topology, 2D topology, 3D topology, etc. ) .
  • non basic tree topology characteristics e.g., characteristics similar to a fat-tree topology, CLOS topology, 2D topology, 3D topology, etc.
  • a general description of method 1100 is presented initially and additional explanation of various aspects of operations (e
  • an apparatus is configured to be associated with a first address.
  • the first address is considered an end-point address by a first device that the apparatus receives information from even though the apparatus forwards the information received from the first device to a second device.
  • the second device is associated with a second address.
  • a first end-point address can be associated with a first communication interface component (e.g., a port of an apparatus, a input/output component of an apparatus, etc. ) .
  • the first end-point address can be associated with at least one communication side of an apparatus (e.g., associated with communications received from an upstream side of a component, etc. ) .
  • the apparatus is configured to be associated with a third address.
  • the third address is considered an end-point address by the second device the apparatus receives information from even though the apparatus forwards the information received from the second device to the first device.
  • the second device is associated with a fourth address.
  • the second address is associated with a second communication interface component.
  • the third address is associated with at least one different communication side of the apparatus (e.g., associated with communications received from a downside communication component, etc. ) .
  • an address translation capability is established.
  • the address translation capability enables translation between the first address and second address and also between the third address and fourth address. Communications received with the first and third addresses are forwarded in accordance with the address translations to the second and fourth addresses respectively, even though the first and third addresses are considered end-point addresses (e.g., in a PCIe compliant topology, etc) .
  • the address translation capability can be compatible with a variety of address allocation and mapping approaches.
  • the translation capability can include a capability of translating between an end-point indication or address that appears as an end-point to at least one device and an indication or address of at least one other device.
  • the apparatus configuration process can facilitate at least one alternative communication path between at least two other devices (e.g., the first device and second device, etc. ) while remaining compatible with a protocol that otherwise restricts or discourages use of an alternative communication path.
  • the alternative communication path between the two devices can be shorter or more direct (e.g., have less hops, etc. ) than another communication path communicatively coupling the two devices.
  • a handshaking communication to establish configuration and initialization of a connection with another device can be performed as part of the apparatus configuration process (e.g., as part of establishing addresses, assignment of communication interfaces, ports or input/output components, etc. ) .
  • Performance of communicatively coupled or networked components is often dependent upon the protocol and topology implemented within a communication environment.
  • Different protocols and topologies often have the potential to offer various performance efficiencies in various environments or applications (e.g., single system, networked systems, servers, data center, etc. ) .
  • Protocols have rules and restrictions that traditionally preclude or discourage implementation with some topologies, and thereby often limit the potential realization of various efficiencies associated with implementation of different protocols’ and topologies’ characteristics and features.
  • an implementation of the new approach can facilitate utilization of efficient protocols in environments with various characteristics (e.g., similar to a variety of topologies, etc. ) that are not otherwise efficiently utilized in conventional approaches.
  • a new apparatus approach facilitates utilization of a PCIe compatible protocol across network configuration similar to various topologies (e.g., fat tree, CLOS, 2D and 3D torus, etc. ) .
  • Some conventional topologies have thresholds or boundaries of diminishing effectiveness where performance begins to be impacted (e.g., greater latency, longer paths, slowed communication, bottlenecks, unduly burdened paths, etc. ) .
  • a new configuration approach facilitates continued efficient operation of a protocol beyond thresholds or boundaries of otherwise diminishing effectiveness.
  • the threshold or boundaries can be associated with various network characteristics (e.g., complexity, size, etc. ) .
  • the thresholds or boundaries can be considered to define domains and the new apparatuses are implemented at the thresholds or boundaries of the domains.
  • the apparatuses can be considered to be implemented at points of otherwise diminishing performance.
  • the apparatuses can enable continued efficient and effective utilization of a protocol beyond points that the protocol would otherwise be less efficient and effective.
  • topology of the components within a domain is relatively simple while the overall configuration topology of multiple domains is considered to be complex and complicated.
  • the new approach can facilitate use of an efficient protocol within a domain and continued efficient use of the protocol across the domains (e.g., unlike conventional attempts at implementing the protocol across the domains which are typically prohibited of often result in diminishing effectiveness) .
  • FIG. 12 is a block diagram of an exemplary network 1200 in accordance with one embodiment.
  • Network 1200 includes domains 1211, 1212, 1213, and 1214 which each include various switches (S) , end-point devices (E) , and root complex components (RC) coupled to each other within a domain in a variety of configurations. Connections within a domain are not shown so as not to obfuscate the flexibility of the illustrated new approach implementation of coupling between the domains. It is appreciated that the switches and end-point devices can be implemented in a variety of configurations.
  • the domains 1211, 1212, 1213, and 1214 can be communicatively coupled by interconnections 1221, 1222 and 1223.
  • the domains 1211, 1212, 1213, and 1214 can also be communicatively coupled by alternative paths that include non-transparent bridge circuits 1251, 1252 and 1253.
  • the communicative coupling of the domains by non-transparent bridge circuits 1251, 1252 and 1253 is compatible with an efficient protocol and implemented without the performance diminishing impacts or characteristics of interconnections 1221, 1222 and 1223.
  • FIG 10 is a flow chart of an exemplary method 1000 for transmitting data packets through a bridge in a PCIe compliant transmission system in accordance with one embodiment.
  • the bridge can include a non-transparent bridge circuit.
  • a processor is initialized to simulate two end-point devices and establish a link between one of the two end-point devices being simulated by the processor and a second switch.
  • the link comprises part of a first data path from the first switch to the second switch.
  • the first PCIe end-point address is mapped to the second address in order to transmit the data packet through the link.
  • the mapping can prevent data packets from being transmitted through a loop transmission path formed at least in part by the first data path and the second data path.
  • the first data path and second data path can enable multi-path routing in the PCIe transmission system.
  • the first path and second path establish a plurality of transmission connections between a plurality of switches, including the first switch and the second switch, to form a high-dimensional topology that is PCIe compliant.
  • a data packet is received in a processor, the data packet originating from a source end-point device from the first switch.
  • the data packet being directed to the first address corresponding to one of the two simulated end-point devices.
  • the first address associated with the bridge e.g., a PCIe end-point address
  • the translation is performed using mapped information from an address translation table.
  • the new configuration approach can enable multi-path routing in a network compatible with PCIe requirements. This can facilitate building more complex network topologies than the basic tree approaches.
  • the new approach can enable realization of many of the characteristics similar to a PCIe protocol while still facilitating realization of characteristics similar to various topologies (e.g., fat-tree, CLOS, 2D or 3D torus, etc. ) .
  • PCIe Non-Transparent Bridges configured in accordance with the new approach can be utilized for interconnecting multiple segments of a topology similar to a tree topology and thus create additional alternative links without breaking or violating the fundamentals of the particular protocol (e.g., PCI, PCIe, etc. ) .
  • PCIe implementation attempts are currently mostly limited to deployment within single enclosures (e.g., servers, PCs, etc. ) and primarily at the printed circuit board (PCB) level
  • PCB printed circuit board
  • the new approach can facilitate implementation of PCIe attributes and advantages in numerous more complex and complicated situations.
  • being compatible with the PCIe protocol enables the new approach to be a flexible and cost efficient alternative to current Data Center interconnect technologies, such as Ethernet (ETH) and InfiniBand (IB) .
  • ETH Ethernet
  • IB InfiniBand
  • a new approach can include many characteristics or features similar to a tree type protocol or topology (e.g., PCI, PCIe, etc. ) .
  • a single link can scale up to 256Gbps with a latency of approximately 130ns per switch hop.
  • the bandwidth of links can be flexibly configured from 1 to 16 lanes (e.g., lane counts can be a power of 2, etc. ) , with up to 8 Giga bits per second (Gbps) per lane (higher throughput may be provided per lane) .
  • the switches are capable of interconnecting links with different speeds (e.g., number of lanes, etc. ) and lane counts per link (e.g., switch port, etc. ) can be dynamically reconfigured.
  • the technology can provide hardware level resource sharing with a large set of different PCIe based I/O devices compatible with this technology.
  • the new non-transparent bridge circuits are capable of interacting with conventional PCIe technology widely used within servers.
  • a CPU can have PCIe compliance built into the chip, not requiring any additional chipset to be connected between the CPU and the peripherals.
  • clusters or Data Centers based on the new approach can be compatible with a PCIe interconnect and achieve greatly improved cost efficiency (e.g., due to the lower number and types of components needed in the system, etc. ) .
  • the power consumption of PCIe switches is several times lower per Gbps, further contributing to the cost efficiency of such a system.
  • a new approach can also provide RDMA (Remote Direct Memory Access) functionality, making it a viable solution for high-speed, low-latency CPU-to-CPU communication.
  • the new approach can also use built-in credit-based flow-control, providing reliable data transfer at the hardware level, requiring a much “slimmer” software stack.
  • the technology can also support cabling for the network to span larger distances and is flexible enough to support different network topologies. There are several solutions with low cost adapters and cables, making it a potential, cost-efficient Data Center interconnect.
  • a new approach can leverage aspects of being compatible with PCIe, including the aspect that the PCIe standardization organization (e.g., PCI-SIG) only defines the protocol and packet formats, but vendors implementing the technology are free to implement any functionality inside their devices (i.e. switches, end-points, etc. ) , providing a great level of flexibility and potential for supporting different functionalities for applications otherwise missing from other Data Center interconnection technologies.
  • PCIe standardization organization e.g., PCI-SIG
  • PCI-SIG PCIe standardization organization
  • vendors implementing the technology are free to implement any functionality inside their devices (i.e. switches, end-points, etc. ) , providing a great level of flexibility and potential for supporting different functionalities for applications otherwise missing from other Data Center interconnection technologies.
  • a new approach compatible with PCI-Express can be identified as a Unified Interconnect for various types of communications among the components within a Data Center.
  • the new approach or methodology facilitates scaling the size of a PCI-Express network beyond its current limitation of 256 nodes, by interconnecting or coupling multiple PCIe network domains.
  • Each PCIe domain has its own management entity or node (e.g., similar to a first PCIe root complex controller, etc. ) that cooperates with other domains’ management entity or node for correct address configuration within its own domain. This allows inter-domain data access between the nodes connected to either network domain.
  • a non-transparent bridge circuit enables scaling.
  • the non-transparent bridge circuit includes a processor coupled to an input port to accept a data packet from a first PCIe network domain comprised of up to 256 nodes; the processor examines an address contained in the data packet and, based on the address, forwards the data packet to a second PCIe network domain comprised of up to 256 nodes, and the processor uses an output port coupled to the processor that transmits the data packet to the second PCIe network domain.
  • the non-transparent bridge circuit is associated with a pair of end-point addresses that are PCIe compliant, wherein a first one of the pair of end-point addresses is assigned by a first PCIe root complex controller associated with the first PCIe network domain and the second one of the pair of end-point addresses is assigned by a second PCIe root complex controller associated with the second PCIe network domain.
  • the data packet is transmitted between the first PCIe network domain and the second PCIe network domain via the non-transparent bridge circuit in accordance with address entries in the address translation table.
  • the non-transparent bridge circuit also includes a memory that stores information and instructions for the processor.
  • the memory can store an address translation table that maps the first one of the pair of end-point addresses to a second address associated with a one of the nodes included in the second PCIe network domain, and maps the second one of the pair of end-point addresses to a fourth address associated with one of the nodes included in the first PCIe network domain.
  • the new approaches described herein can facilitate high-dimensional PCI-Express (PCIe) network implementations.
  • the new approaches can facilitate utilization of an efficient protocol (e.g., PCIe, etc. ) while enabling implementation of various characteristics and features (e.g., characteristics and features similar to a fat-tree topology, CLOS topology, 2D and 3D topologies, etc. ) that would otherwise not be compatible with the protocol.
  • an efficient protocol e.g., PCIe, etc.
  • characteristics and features e.g., characteristics and features similar to a fat-tree topology, CLOS topology, 2D and 3D topologies, etc.
  • alternative paths can be enabled and utilized while maintaining compliance with a protocol (e.g., PCIe, etc. ) that would otherwise not be compatible with the use of alternative paths.
  • the alternative paths can facilitate flexible topology implementation and network domain scaling while enabling improved communication latency.
  • presented systems and methods facilitate utilization of a non-transparent bridge circuit configured as an end-
  • Embodiments described herein may be discussed in the general context of computer-executable instructions, such as program modules, residing on some form of computer-readable storage medium executed by one or more computers or other devices.
  • computer-readable storage media may comprise non-transitory computer-readable storage media.
  • Non-transitory computer-readable storage media includes all computer-readable media except for a transitory, propagating signal.
  • Computer-readable storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules or other data.
  • program modules include routines, programs, objects, components, data structures, etc., that perform particular tasks or implement particular abstract data types. The functionality of the program modules may be combined or distributed as desired in various embodiments.

Landscapes

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

Abstract

Les descriptions présentées dans l'invention comprennent la mise en oeuvre de configurations de réseau PCI-Express (PCIe) de grandes dimensions. Ces nouvelles approches peuvent faciliter l'utilisation d'un protocole efficace (p. ex. PCIe, etc.) et permettent en même temps de mettre en oeuvre divers attributs et caractéristiques (par ex. attributs et caractéristiques similaires à une topologie en arbre élargi, à une topologie CLOS, à des topologies 2D et 3D, etc.) qui ne seraient par ailleurs pas compatibles avec le protocole. Par exemple, la mise en oeuvre de trajets alternatifs peut être activée et utilisée tout en maintenant la conformité avec un protocole (p. ex. PCIe, etc.) qui ne serait par ailleurs pas compatible avec l'utilisation de trajets alternatifs. Les trajets alternatifs peuvent faciliter la mise en oeuvre d'une topologie et d'une mise à l'échelle de domaine de réseau flexibles tout en améliorant la latence de communication. Dans un mode de réalisation, les systèmes et procédés présentés facilitent l'utilisation d'un circuit en pont non transparent, configuré sous la forme d'un point d'extrémité par rapport à des communications provenant d'au moins un dispositif, et permettent en même temps de transmettre des communications sur au moins un autre dispositif.
PCT/CN2015/093802 2014-11-07 2015-11-04 Procédé et appareil de pont non transparent pour configurer des réseaux pci-express de grandes dimensions WO2016070813A1 (fr)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN201580059369.2A CN107113210B (zh) 2014-11-07 2015-11-04 用于配置高阶快捷pci网络的非透明桥式方法和装置
EP15856318.9A EP3172870B1 (fr) 2014-11-07 2015-11-04 Procédé et appareil de pont non transparent pour configurer des réseaux pci-express de grandes dimensions

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US201414145365A 2014-11-07 2014-11-07
US14,536/516 2014-11-07

Publications (1)

Publication Number Publication Date
WO2016070813A1 true WO2016070813A1 (fr) 2016-05-12

Family

ID=55908598

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2015/093802 WO2016070813A1 (fr) 2014-11-07 2015-11-04 Procédé et appareil de pont non transparent pour configurer des réseaux pci-express de grandes dimensions

Country Status (1)

Country Link
WO (1) WO2016070813A1 (fr)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3503483A1 (fr) * 2017-12-21 2019-06-26 Bull Sas Procédé et serveur d'attribution d'adresses topologiques à des commutateurs de réseau, programme d'ordinateur et grappe de serveurs correspondants

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090164694A1 (en) * 2007-12-21 2009-06-25 Aprius Inc. Universal routing in pci-express fabrics
US20090198863A1 (en) * 2006-10-10 2009-08-06 International Business Machines Corporation Transparent pci-based multi-host switch
WO2009120798A2 (fr) * 2008-03-25 2009-10-01 Aprius Inc. Serveur mandataire de fonction pci-express
US8429325B1 (en) * 2010-08-06 2013-04-23 Integrated Device Technology Inc. PCI express switch and method for multi-port non-transparent switching

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090198863A1 (en) * 2006-10-10 2009-08-06 International Business Machines Corporation Transparent pci-based multi-host switch
US20090164694A1 (en) * 2007-12-21 2009-06-25 Aprius Inc. Universal routing in pci-express fabrics
WO2009120798A2 (fr) * 2008-03-25 2009-10-01 Aprius Inc. Serveur mandataire de fonction pci-express
US8429325B1 (en) * 2010-08-06 2013-04-23 Integrated Device Technology Inc. PCI express switch and method for multi-port non-transparent switching

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3503483A1 (fr) * 2017-12-21 2019-06-26 Bull Sas Procédé et serveur d'attribution d'adresses topologiques à des commutateurs de réseau, programme d'ordinateur et grappe de serveurs correspondants
FR3076142A1 (fr) * 2017-12-21 2019-06-28 Bull Sas Procede et serveur d'attribution d'adresses topologiques a des commutateurs de reseau, programme d'ordinateur et grappe de serveurs correspondants
US20190215242A1 (en) * 2017-12-21 2019-07-11 Bull Sas Method and server for assigning topological addresses to network switches, corresponding computer program and server cluster
US11005714B2 (en) 2017-12-21 2021-05-11 Bull Sas Method and server for assigning topological addresses to network switches, corresponding computer program and server cluster

Similar Documents

Publication Publication Date Title
US9647962B2 (en) Non-transparent bridge method and apparatus for configuring high-dimensional PCI-express networks
US9043526B2 (en) Versatile lane configuration using a PCIe PIe-8 interface
US9294403B2 (en) Mechanism to control resource utilization with adaptive routing
RU2543558C2 (ru) Способ и усройство маршрутизации ввода-вывода и карта
US20140092740A1 (en) Adaptive packet deflection to achieve fair, low-cost, and/or energy-efficient quality of service in network on chip devices
KR101559644B1 (ko) 통신 제어 시스템, 스위치 노드 및 통신 제어 방법
US9025495B1 (en) Flexible routing engine for a PCI express switch and method of use
US10263891B2 (en) Switching device, computer system, method, and program
CN101242371B (zh) 基于PCIe交换架构路由器堆叠的方法、系统和装置
US8463977B2 (en) Use of PCI express for CPU-to-CPU communication
US20150261709A1 (en) Peripheral component interconnect express (pcie) distributed non- transparent bridging designed for scalability,networking and io sharing enabling the creation of complex architectures.
US10587535B2 (en) Adding a network port to a network interface card via NC-SI embedded CPU
US10191882B2 (en) Method and system for aggregation-friendly address assignment to PCIe devices
CN108021525B (zh) 一种基于pcie总线多主互连的冗余交换系统
WO2024212744A1 (fr) Serveur, dispositif hétérogène et appareil de traitement de données associé
US20240020261A1 (en) Peer-to-peer route through in a reconfigurable computing system
WO2016070813A1 (fr) Procédé et appareil de pont non transparent pour configurer des réseaux pci-express de grandes dimensions
US20050038949A1 (en) Apparatus for enabling distributed processing across a plurality of circuit cards
KR20150102538A (ko) 시스템 온칩 통신 네트워크
Fukase et al. A study on Inter-Board Data Transmission of FPGA Cluster
US20240073129A1 (en) Peer-to-peer communication between reconfigurable dataflow units
CN207022032U (zh) 一种业务线卡及基于pcie总线背板的通信系统
CN115701060A (zh) 一种报文传输方法及相关装置
US20130268684A1 (en) Device for multiple ethernet connection to a computer unit and computer unit assembly and equipment linked together
Kaushik et al. Available Online at www. ijesar. in

Legal Events

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

Ref document number: 15856318

Country of ref document: EP

Kind code of ref document: A1

REEP Request for entry into the european phase

Ref document number: 2015856318

Country of ref document: EP

WWE Wipo information: entry into national phase

Ref document number: 2015856318

Country of ref document: EP

NENP Non-entry into the national phase

Ref country code: DE