Fault tolerance

Abstract

A configurable switching fabric port is disclosed having, in a particular configuration. A first interface that employs port interface resources and leaves at least one interface resource dormant and a second interface utilizing the dormant resource. One particular fault non-tolerant architecture, the RapidIO System, is specifically addressed. One implementation of this system incorporates transmission and reception ports configurable as 16 and 8 bit interfaces. In the 8-bit configuration, an 8-bit interface incorporates the least significant 8-bits of signal resources. Further, in the reduced, or 8-bit configuration, the most significant port interface resources of the 16 bit port are surplus.

Description

FIELD OF THE INVENTION

[0001] The present invention relates to fault tolerance, and more particularly to fault tolerance in data networks using a point to point, packet switched, fabric architecture.

BACKGROUND OF THE INVENTION

[0002] It is the nature of the computer system industry to require an exponential performance advantage over the generations while maintaining or decreasing system costs. In particular, telecommunications and networking systems benefit from a reduction in system size and an increase in capabilities.

[0003] Therefore, a point to point, packet switched, fabric architecture is displacing traditional memory mapped bus architecture for use in network equipment, storage subsystems and computing platforms capable of providing an interface for processors, memory modules and memory mapped I/O devices.

[0004] Modern digital data networks are increasingly employing such point to point, packet switched, fabric interconnect architectures to overcome bandwidth limitations. These networks transmit encapsulated address, control and data packets from the source ports across a series of routing switches or gateways to addressed destinations. The switches and gateways of the switching fabric are capable of determining from the address and control contents of a packet, what activities must be performed.

[0005] Incorporating a level of fault tolerance in a packet switched network is highly desirable. Fault tolerance is the ability of a system to respond gracefully to an unexpected component failure. Traditionally, fault-tolerance has referred to building subsystems from redundant components that are placed in parallel; Faults are determined above the physical level of the protocol based on communication failure; such information is relayed to the physical layer, which can employ redundancy. Failure to account for faults will render at least that port inoperative, which may result in larger scale, possibly system wide failure, depending on the nature of the component corresponding to the port.

[0006] There are a number of architectures for proving fault tolerance. These architectures can be grouped into cold, warm and hot standby, or load shared. Cold stand-by refers to equipment that can be started once the first unit fails.

[0007] Dead time will occur while the replacement unit is started, switched into place, and lost data is retransmitted. Warm stand-by refers to equipment that is always running pending failure of the first unit. A shorter dead time will occur while the second unit is switched into the first unit's place, and lost data is retransmitted. Hot stand-by refers to equipment that is always running, and is always hooked up ready to take over if the first unit fails. Hot standby equipment does not actually carry any traffic until the first unit fails, but no dead time interrupts communications when the first unit fails. Load Shared refers to equipment that is always running, and is always hooked up transmitting data in combination with the primary unit.

[0008] In order to ensure compatibility, fabric architectures must adhere to standards. Introduction of additional features in standard compliant systems requires the implementation of such features to be adapted to standard requirements of the existing architecture.

[0009] It is, therefore, necessary in implementing the point-to-point, packet-switched architecture described above, to consider the level of fault tolerance mandated for the system to which it is directed. Where fault tolerance is required, but not provided for by a standard, the system must have a method and/or an apparatus to overcome failure.

[0010] In the instance of switching fabrics, should an individual interface fail to communicate with the fabric, it is desirable for the interface to redundantly connect to an alternate fabric. However, a redundant interface dedicated to an alternate fabric would require a full complement of interface resources to implement. This failure could occur in the port, the fabric, or on the printed circuit board connecting the two.

[0011] It is often the case that configuration circumstances leave resources dormant in particular configurations.

[0012] In one standard, RapidIO System, a physical specification is defined (RapidIO Interconnect Specification Part IV: Physical Layer 8/16 LP-LVDS Specification) with the flexibility to support dedicated 8 or 16 bit interfaces. Where a RapidIO port has been designed to be configurably connectable to either standard bus, but is only using the 8 bit configuration, some signal resources are left idle. The RapidIO standard is compatible with cold and hot standby and provides for guaranteed message delivery.

[0013] In another standard, HyperTransport™ I/O Link Specification (Revision 1.03), a protocol is defined with the flexibility to support dedicated 2, 4, 8, 16 or 32 bit interfaces. Utilized width is accomplished by negotiating a link compatible with the smallest end. As in the case of RapidIO, some signal resources are left idle in non-32 bit configurations. The HyperTransport™ standard is compatible with cold standby and does not provide for guaranteed message delivery.

[0014] What is needed is a fault tolerant adaptation of existing architectures that minimizes additional resources required to support redundancy.

SUMMARY OF THE INVENTION

[0015] It is an advantage of the disclosed invention to adapt dormant resources of an existing fault non-tolerant architecture in order to provide for a fault tolerance mode (whether it be cold or hot stand by).

[0016] In a corresponding embodiment a configurable switching fabric port is disclosed having, in a particular configuration:

[0017] A first interface that employs port interface resources and leaves at least one interface resource dormant.

[0018] And a second interface utilizing the dormant resource.

[0019] One particular fault non-tolerant architecture, the RapidIO System, is specifically addressed. One implementation of this system incorporates transmission and reception ports configurable as 16 and 8 bit interfaces.

[0020] In the 8-bit configuration, an 8-bit interface incorporates the least significant 8-bits of signal resources. Further, in the reduced, or 8-bit configuration, the most significant port interface resources of the 16 bit port are surplus.

[0021] It is an advantage of the disclosed invention to adapt these surplus resources in order to provide for a redundant interface.

[0022] A second fault non-tolerant architecture, HyperTransport™ is also addressed. The HyperTransport™ standard supports interfaces that are configurable to 2, 4, 8, 16, or 32 bits in width. Wider links connected to narrower links negotiate to the least common width. In this case also, port resources on the wider link are left surplus, and available for enhancing system fault tolerance.

[0023] In corresponding embodiments of the invention, a second interface is provided incorporating those surplus most significant byte resources.

BRIEF DESCRIPTION OF DRAWINGS

[0024] The present invention will be further understood from the following detailed description with reference to the drawings in which:

[0025] FIG. 1 is a schematic of an 8 bit RapidIO Network;

[0026] FIG. 2 is a schematic of a 16 bit RapidIO Network;

[0027] FIG. 3 is a partial schematic of a RapidIO Network with 8 and 16 bit portions;

[0028] FIG. 4 is a schematic of a RapidIO device capable of communication over an 8 bit bus or a 16 bit bus;

[0029] FIG. 5 is a schematic in detail of a transmission port capable of communication over 8 or 16 bit busses;

[0030] FIG. 6 is a schematic in detail of a reception port capable of communication over 8 or 16 bit busses;

[0031] FIG. 7 is a schematic of a RapidIO device capable of communication over one or two 8 bit busses or one 16 bit bus;

[0032] FIG. 8 is a schematic in detail of a transmission port capable of communication over one or two 8 bit busses or one 16 bit bus;

[0033] FIG. 9 is a schematic in detail of a reception port capable of communication over one or two 8 bit busses or one 16 bit bus; and

[0034] FIG. 10 is a schematic of a fault tolerant network incorporating an embodiment of the present invention.

DETAILED DESCRIPTION

[0035] Existing topologies provide for configuration of ports to adapt to multiple bus widths: FIG. 1 shows a network 1 with a device 10 and a device 20 connected by an 8 bit RapidIO Bus 30. FIG. 2 shows a network 2 with the same device 10 and another device 25 connected by an 16 bit RapidIO Bus 40. FIG. 3 shows a portion of a network 3 comprising a switch 50 connecting 28 bit busses 60, 70 and a 16 bit bus 80.

[0036] The devices require RapidIO physical ports in order to connect to their busses. Device 10 requires a configurable port 90 in order to provide for connection with either bus 30 or 40. It is commercial advantageous to produce a port with such flexibility. Device 20 employs a dedicated 8 bit port 92 and device 25 employs a dedicated 16 bit port 94. Device 50 employs 3 configurable ports 90. Since device 50 employs the configurable ports 90, it can bridge a variety of three bus situations. Knowledge of the art will suggest any number of variations of combinations of these 3 types of ports.

[0037] In FIG. 4 we show greater detail of a device 100 capable of utilization in multiple configurations: Device 100 is connected to a RapidIO network. The device 100 is connected to the network by a bus 108. The port 90 is of the configurable type; bus 108 may be 8 or 16 bit. The port connects immediately with a device core 104; the core 104 comprises any RapidIO physical layer functions not comprised by port 90 as well as the functions of the transport layer and higher levels.

[0038] FIGS. 5 and 6 detail the port 90 and it's immediate circumstances. Taking the half-ports 90a and 90b of FIGS. 5 and 6 together, we have the complete port 90.

[0039] FIG. 5 shows greater detail of a portion of the device 100 consisting of a transmission portion 90a of a known RapidIO transmission port 90 and the core 104. Connected to the core 104 are two 8-bit busses 110 and 120. Bus 110 is directly connected to a least significant byte (LSB)/8 bit output port 130.

[0040] Bus 120 is coupled to a most significant byte (MSB) port 140 by a buffer 150, in turn coupled to the core 104 by a 16 bit mode enable signal path 160. Coupled directly from the core is a frame signal path 170, and 2 clock signal paths 180, 190. Note that under the standard, there are two output ports for the clock signal, and each signal line is a low voltage differential pair.

[0041] Bus 110 asserts a least significant portion of a 16 bit datum or the entirety of an 8 bit datum. Bus 120 asserts a most significant portion of a 16 bit datum. The frame signal exists to communicate the intervals of 32 bit datum corresponding to groups of 8 or 16 bit transmissions. The core controls the configuration of the half-port 90a via the 16 bit mode enable signal path 160.

[0042] FIG. 6 shows greater detail of a portion of the device 100 consisting of a reception portion 90b of a known RapidIO transmission port 90 and the core 104. Connected to the core 104 are two 8-bit busses 210 and 220. Bus 210 is directly connected to a LSB/8 bit input port 230. Bus 220 is coupled to a MSB port 240 by a buffer 250, in turn coupled to the core 104 by a 16 bit mode enable signal path 260. Coupled directly from the core 104 is a frame signal 270 and 2 clock signals 280, 290. Note that under the standard, there are two input ports for the clock signal, and each signal line is a low voltage differential pair.

[0043] Bus 210 inputs the least significant portion of a 16 bit datum or the entirety of an 8 bit datum. Bus 220 inputs a most significant portion of a 16 bit datum. The frame signal exists to communicate the intervals of 32 bit datum corresponding to groups of 8 or 16 bit transmissions. The core controls the configuration of the half-port 90b via the 16 bit mode enable signal path 260.

[0044] Taken together, port 90a and 90b form the complete port 90. This ports is configurable via signals 160 and 260 as an 8 bit interface (bus 110 for transmission/bus 210 for reception) or a 16 bit interface (bus 110 plus 120 for transmission/bus 210 plus 220 for reception).

[0045] An embodiment of the present invention can be seen in FIGS. 7a and 7b, which shows a device 100′ equipped with the innovation as it may be connected in two RapidIO networks. The improved port 90′ connects immediately with a device core 104′; the core 104′ comprises any RapidIO physical layer functions not comprised by improved port 90′ as well as the functions of the transport layer and higher levels. In the first deployment, shown in FIG. 7a, the device 100′ is connected to the network by a bus 108. As the improved port 90′ is of the configurable type, bus 108 may be 8 or 16 bit. In the second deployment, shown in FIG. 7b, the device 100′ is connected to the network by two 8 bit busses 109a and 109b. In order for the improved port 90′ to be compatible with these two deployments, modifications, differing from port 90, are required.

[0046] FIGS. 8 and 9 detail the improved port 90′ and it's immediate circumstances. Taking the half-ports 90a′ and 90b′ of FIGS. 8 and 9 together, we have the complete improved port 90′.

[0047] FIG. 8 shows a portion of the device 100′ consisting of a transmission portion 90a′ of the improved RapidIO transmission port 90′ and the core 104′. Connected to the core 104′ are two 8-bit busses 310 and 320. Bus 310 is directly connected to a LSB/8 bit output port 330 and coupled to a MSB port 340 by a buffer 332. Bus 320 is coupled to the MSB port 340 by a buffer 350, in turn coupled to the core 104′ by a 16 bit mode enable signal path 360. Coupled directly from the core is a frame signal path 370, and two clock signal paths 380, 390. A duplicate frame signal path 370′ is coupled to the frame signal path 370 by a buffer 382. Buffers 332 and 382 are coupled to a fault mode control 336 by a fault/8 signal path 334. The fault mode control 336 is coupled to the core by the 16 bit mode enable signal path 360 and by the fault signal path 362. Note that under the standard, there are two output ports for the clock signal, and each signal line is a low voltage differential pair.

[0048] In operation, bus 310 asserts a least significant portion of a 16 bit datum or the entirety of an 8 bit datum. Bus 320 asserts a most significant portion of a 16 bit datum. The frame signal exists to communicate the intervals of 32 bit datum corresponding to groups of 8 or 16 bit transmissions. The core controls the configuration of the half-port 90a′ via the 16 bit mode enable signal path 360 and the fault signal path 362. RapidIO busses may be connected to 90a′ in the following formats: one 8 bit bus to signals 330, 370, 380 or to 340, 370′, 390; One 16 bit bus to signals 330, 340, 370, 380; or two 8 bit busses, one to 330, 370, 380, the other to 340, 370′, 390. This last format is ideally suited for fault tolerant swapping between busses, as directed by core 104′, and described herein below.

[0049] FIG. 9 shows a portion of the device 100′ consisting of a reception portion 90b′ of the improved RapidIO transmission port 90′ and the core 104′. Connected to the core 104′ are two 8-bit busses 410 and 420. Bus 410 is directly connected to a LSB/8 bit output port 430 and coupled to a MSB port 440 by a buffer 432. Bus 420 is coupled to the MSB port 440 by a buffer 450, in turn coupled to the core 104′ by a 16 bit mode enable signal path 460. Coupled directly from the core is a frame signal path 470, and a two clock signal paths 480, 490. A duplicate frame signal path 470′ is coupled to the frame signal path 470 by a buffer 482. Buffers 432 and 482 are coupled to a fault mode control 436 by a fault/8 signal path 434. The fault mode control 436 is coupled to the core by the 16 bit mode enable signal path 460 and by the fault signal path 462. Note that under the standard, there are two output ports for the clock signal, and each signal line is a low voltage differential pair.

[0050] Bus 410 asserts a least significant portion of a 16 bit datum or the entirety of an 8 bit datum. Bus 420 asserts a most significant portion of a 16 bit datum. The frame signal exists to communicate the intervals of 32 bit datum corresponding to groups of 8 or 16 bit transmissions. The core controls the configuration of the half-port 90b′ via the 16 bit mode enable signal path 460 and the fault signal path 462. RapidIO busses may be connected to 90b′ in formats complementary to those of 90a′.

[0051] Taken together, port 90a′ and 90b′ form the complete port 90′. The port 901 is configurable via signals 360 and 460 as a 16 bit interface (bus 310 plus bus 320 for transmission/bus 410 plus bus 420 for reception). The port 901 is also configurable via signals carried on signal paths 160 and 260 as two different 8 bit interfaces. Signals carried on signal paths 362 and 462 determine whether such an interface is formed with bus 310 or 320 for transmission/bus 410 or 420 for reception). In a fault tolerant system, the port 901 provides for the necessity of rerouting (of 8 bit signals).

[0052] FIG. 10 details a network of an embodiment of the present innovation. Three network endpoints, 500a, 500b, and 500c are serviced by the network. These devices are of a class compliant with device 100′ i.e. having an improved port 90′ compliant port, 501a, 501b, and 501c respectively. The network also includes two switches, 600a and 600b. Each of these switches incorporates 3 ports, 601a, 602a, 603a, 601b, 602b, and 603b respectively. Each endpoint 500a, 500b, 500c is connected to the primary network switch 600a by a RapidIO Bus, 510a, 510b, and 510c, respectively. This connection is made to the LSB of ports 501a, 501b, and 501c respectively, and to 601a, 602a, 603a respectively.

[0053] Each endpoint 500a, 500b, 500c is connected to the replacement network switch 600b by a RapidIO Bus, 520a, 520b, and 520c, respectively. This connection is made to the MSB of ports 501a, 501b, and 501c respectively, and to 601b, 602b, 603b respectively.

[0054] In the RapidIO system, IDLE communications are continuously transmitted in the absence of significant communications. This forms a discernable ‘heartbeat’. When failure is detected, through the absence of the heartbeat, and provided any request for re-training fails, a fault tolerance routine may be executed.

[0055] For example, if switch 600a, or bus 510a fails, switch 600b is notified to ready for communication, the busses 520a, 520b, and 520c are trained, data is recovered for packets lost in switch 600a (RapidIO guarantees message delivery), and same switch is notified to terminate communication. Switch 600b resumes the function of 600a. Note that the standby mode in which the switch 600b is maintained (eg. Hot, Cold) is dependent on the actions and response implemented at a higher level of protocol.

[0056] It can be understood by one skilled in the art that the aforementioned RapidIO adaptations are equally applicable to HyperTransport™ or a similar standard, and the innovation is not derived from the standards but applicable to such.

[0057] For example, the mechanism described for overlaying this fault tolerance enhancement on the RapidIO standard is equally applicable to the HyperTransport™ interface.

[0058] In the case of a HyperTransport™ interface, there are several signals defined which are different from RapidIO. One skilled in the art could identify that these signals (e.g. CTL, PWROK, RESET#, LDTSTOP#, LDTREQ#) could be replicated in the same way as the FRAME signal in the above RapidIO examples. While these signals perform different functions than the FRAME signal of RapidIO, one skilled in the art can understand that the method for replicating these signals across several fault tolerant interfaces is the same as RapidIO. In the case of HyperTransport™, where there is a wider choice of data bus widths available the multiplicity of redundant interfaces could be correlated with the un-used data lines.

Claims

1. A port for connecting to a switching fabric, having a plurality signal resources dividable into at least a first resource subset and a second resource subset, said port configurable in a plurality of configurations and said port comprising: a first interface operable whereby said port is in a first configuration, said first interface incorporating said first resource subset and, said first interface not incorporating said second resource subset, a second interface operable whereby said port is in a second configuration, said second interface incorporating said first resource subset and said second resource subset, a third interface operable whereby said port is in a third configuration, said third interface incorporating said second resource subset and, said third interface not incorporating said first resource subset.

2. The port of claim 1, for co-operating with a core, and additionally comprising a switching circuit wherein said switching circuit is controlled by the core and capable of coupling either said first, second, or third interface to said core.

3. A method of fault tolerance in a network having a primary fabric a replacement fabric and an endpoint, said endpoint including a port having a plurality signal resources dividable into at least a first resource subset and a second resource subset, said port configurable in a plurality of configurations

4 A method as claimed in claim 3, where the switching fabric comprises a first fabric and a second fabric.

5 A method as claimed in claim 4, wherein the first fabric is a primary fabric.

6 A method as claimed in claim 5, wherein the second fabric is a replacement fabric.

7 A method as claimed in claim 6, wherein the replacement fabric comprises a cold standby fabric.

8 A method as claimed in claim 6, wherein the replacement fabric comprises a hot standby fabric.

9 A method as claimed in claims 1-7, wherein the port is compliant with a standard.

10 A method as claimed in claim 9, wherein the standard is RapidIO.

11 A method as claimed in claim 9, wherein the standard is Hpyertransport™.