Advanced switching architecture

Abstract

The present invention provides a system and method for encapsulating protocols across a switching fabric network.

Description

BACKGROUND OF THE INVENTION

[0002] 1. Field of the Invention

[0003] This invention generally relates to the field of networks, in particular, an extensible switching fabric framework for encapsulation of a protocol.

[0004] 2. Description of the Related Art

[0005] Switch fabric networks are generally known to one skilled in the art. An example of one is described in U.S. patent application Ser. No. 10/152,656 filed May 21, 2002 which is incorporated by reference herein in its entirety.

[0006] In a computer system, the interconnection architecture of various components greatly affect the speed and cost of the system. The PCI standard (i.e., “Peripheral Component Interconnect”), was originally developed as a way to connect different peripherals to a computer.

[0007] Yet another architecture is PCI Express which is a serial interconnect architecture. PCI Express is also used in a switch fabric network architecture. The PCI Express standard retains the PCI usage and software interfaces thereby providing the ability to be compatible with legacy computers (such as a personal computer, “PC”) hardware and software infrastructure. However, this limits the potential of the standard because of constraints from PC architectures retained from the 1980s. Accordingly, there is a need to address the limitations of the PCI Express standard and to provide a way to improve the flexibility and expandability of the PCI Express standard.

BRIEF SUMMARY OF THE INVENTION

[0008] The present invention is an advanced switching architecture that provides extensions to, preferably, the PCI Express standard, as further discussed below.

BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The figures below depict various aspects and features of the present invention in accordance with the teachings herein.

[0010] FIGS. 1-3 and 5-17 depict different data formats for the protocol described herein.

[0011] FIG. 4 depicts the turn pool reversal in accordance with one aspect of the invention.

DETAILED DESCRIPTION OF THE INVENTION

[0012] In accordance with one aspect of the present invention, the advanced switch (“AS”) architecture, provides for an extensible switching fabric framework for encapsulation of virtually any protocol. Preferably, as applied to PCI Express, the PCI Express Advanced Switching (PCI “ExAS”) architecture partitions protocol encapsulation into two distinct categories:

[0013] ExAS Routing; and

[0014] Protocol Encapsulation Interface, which is protocol specific.

[0015] 1. ExAS Routing Header

[0016] In accordance with one aspect of the present invention, the separation of all routing associated information enables simple, high performance and cost effective switch designs that are agnostic to everything within the packet with the exception of what is needed to forward a packet towards the next hop within the fabric. The information that is contained within an ExAS route header includes:

[0017] Path routing information

[0018] Traffic Class ID

[0019] Congestion and Deadlock avoidance information

[0020] Packet Size

[0021] Protocol Encapsulation Interface specification

[0022] a. Unicast Packet Format

[0023] A unicast packet is an encapsulation packet type. The unicast packet format is shown in FIG. 1 and further described in Table 1.

TABLE 1
Unicast Packet Field Description
Bit	Bit field	Hop
Positions	width	Mutable	Field Name	Description
7:0	8	No	Protocol	The interface field indicates the absolute packet type of the
			Encapsulation	encapsulated packet (interfaces 0-191) or the implementation
			Interface	specific packet type of the encapsulated packet (interfaces
				192-254). Interface 255 is reserved.
13:8	6	No	Credit Length	The credit length field does not specify the exact length of the
				associated encapsulated packet. Instead in indicates the
				amount of credit consumed by the encapsulated packet. The
				start and end commas of the link layer determine the exact
				length of the packet. ExAS packet credit is consumed in 32-
				byte blocks. The credit length field indicates the number of
				32-byte blocks necessary to hold the ExAS header and
				encapsulated packet. For example, a packet that at the link
				layer consumes 68 bytes (1-byte start comma, 2 byte sequence
				number, 8-byte ExAS header, 52-byte encapsulated packet, 4-
				byte LCRC, 1-byte end comma), would require two 32-byte
				storage blocks (the encapsulated packet plus header occupy
				only 60-bytes). Similarly, an 80-byte link layer packet would
				require three 32-byte blocks (the encapsulated packet plus
				ExAS header consume 76 bytes, 4 more bytes than could be
				stored in only two 32-byte blocks).
				Bit 5 of the credit length field is the bias bit. When bit 5 is
				clear (0), the lower 5 bits of the field indicate the number of
				32-byte blocks of storage required to forward the packet, from
				0 to 31. A zero length packet indicates that there is no
				encapsulated packet present. When bit 5 is set, the lower 5
				bits of the credit length field contain the number of 128-byte
				blocks (four 32-byte blocks) of storage consumed by the
				encapsulated packet plus 1024-bytes. The equation for
				determining the number of 32-byte credit blocks consumed by
				an ExAS packet's encapsulated packet is:
				blocks = (CL [5]) ? ((CL[4:0] + 8) << 2) : CL[4:0]
16:14	3	No	Traffic Class	This field indicates the priority of the packet. It is used by
				nodes to determine the order in which packets are forwarded
				or processed.
17	1	No	Bypass	This flag is used purely for packet by-pass and flow-control
			Queue	crediting purposes. This bit indicates whether a given packet
				consumes posted/completion credit and is not by-passable (bit
				is clear), or whether the packet consumes non-posted credit
				and is bypassable (bit is set). This bit does not necessarily
				indicate anything about the type of the packet encapsulated by
				the ExAS header. Care should be taken to ensure that this bit
				is properly set when injecting packets into the fabric.
				Improper use of this bit can result in fabric deadlock.
18	1	Yes	Credit Type	This bit indicates whether the associated packet has been
				forwarded using next turn credit or traffic class credit.
24:19	6	Yes	Bit Count	This field indicates the current position of return path head
				within the turn pool. This field also differentiates a unicast
				packet from a multicast packet. Field values between 63 and
				23 are interpreted as unicast packets. Packets that are received
				by switches with bit count values of 23 and 24 indicate that
				the packet has an invalid turn pool (software has improperly
				initialized whatever hardware structure set the packet's turn
				pool). Packets that are received by endpoints with bit count
				values of 23 indicate that the packet has an invalid turn pool.
63:25	39	Yes	Turn Pool	The turn pool contains the variable bit width turn values of a
				path specification.

[0024] b. Multicast Packet Format

[0025] A multicast packet is an encapsulated packet type. Multicast packets cannot use turn credit and always consume traffic class credit. Multicast packets are write only. There is no option for multicast packet bypass. The multicast packet format is shown in FIG. 2 and further described in Table 2.

TABLE 2
Multicast Packet Field Descriptions
Bit	Bit field	Hop
Positions	width	Mutable	Field Name	Description
7:0	8	No	Protocol	The interface field indicates the absolute packet type of the
			Encapsulation	encapsulated packet (interfaces 0-191) or the implementation
			Interface	specific packet type of the encapsulated packet (interfaces
				192-254). Interface 255 is reserved.
13:8	6	No	Credit Length	The credit length field does not specify the exact length of the
				associated encapsulated packet. Instead in indicates the
				amount of credit consumed by the encapsulated packet. The
				start and end commas of the link layer determine the exact
				length of the packet. ExAS packet credit is consumed in 32-
				byte blocks. The credit length field indicates the number of
				32-byte blocks necessary to hold the ExAS header and
				encapsulated packet. For example, a packet that at the link
				layer consumes 68 bytes (1-byte start comma, 2 byte sequence
				number, 8-byte ExAS header, 52-byte encapsulated packet, 4-
				byte LCRC, 1-byte end comma), would require two 32-byte
				storage blocks (the encapsulated packet plus header occupy
				only 60-bytes). Similarly, an 80-byte link layer packet would
				require three 32-byte blocks (the encapsulated packet plus
				ExAS header consume 76 bytes, 4 more bytes than could be
				stored in only two 32-byte blocks).
				Bit 5 of the credit length field is the bias bit. When bit 5 is
				clear (0), the lower 5 bits of the field indicate the number of
				32-byte blocks of storage required to forward the packet, from
				0 to 31. A zero length packet indicates that there is no
				encapsulated packet present. When bit 5 is set, the lower 5
				bits of the credit length field contain the number of 128-byte
				blocks (four 32-byte blocks) of storage consumed by the
				encapsulated packet plus 1024-bytes. The equation for
				determining the number of 32-byte credit blocks consumed by
				an ExAS packet's encapsulated packet is:
				blocks = (CL [5]) ? ((CL[4:0] + 8) << 2) : CL[4:0]
16:14	3	No	Traffic Class	This field indicates the priority of the packet. It is used by
				nodes to determine the order in which packets are forwarded
				or processed.
18:17	2	No	Reserved
24:19	6	Yes	Hop Count	This field indicates the number of switch traversals the
				associated packet has undergone. This field also differentiates
				a unicast packet from a multicast packet. Field values between
				0 and 22 are interpreted as unicast packets. Packets that are
				received by switches with a bit count values of 22 indicate
				that software has improperly initialized the associated
				multicast group.
31:25	7	No	Reserved
51:32	20	No	Multicast	The multicast group manager assigns this field to an endpoint.
			Source Index	It is optional from a routing perspective. That is, multicast
				distribution is unaffected by the value of this field, but it is
				strongly recommended that the multicast group manager
				provide every endpoint with a unique source index.

[0026] 2. Protocol Encapsulation Interface

[0027] The Protocol Encapsulation Interface (“PEI”) allows for protocol encapsulations to include whatever else they need in a subsequent sub-header, or sub-headers, enabling the efficiencies of a modular approach to protocol encapsulation.

[0028] 3. PCI Express-AS Packet Routing

[0029] The ExAS protocol supports two mechanisms for routing packets across an ExAS fabric: path routing (PR) and multicast routing (MR). Path routed packets rely on an explicit source supplied path for routing unicast packets between nodes. Multicast routing provides a true multicast capability for writing data across a fabric based on a table look-up at each node.

[0030] ExAS unicast packet routing is based on an origin and a terminus. The origin of a packet is the endpoint that generates the initial fabric transaction, and the terminus of a packet is the endpoint that ultimately receives the transaction. Endpoints do not forward/switch ExAS packets back into the ExAS fabric on which they arrived.

[0031] a. Path Routing

[0032] All ExAS nodes are required to support path routing. A path specifies the position of the terminus relative to the origin, and is assigned to the ExAS header by the origin of the packet. Nodes are required only to forward the packet according to the path that is contained in the ExAS packet header.

[0033] i. Path Specification

[0034] A path, as shown, for example, in FIGS. 1 and 3, consists of a bit count, and a turn pool. The bit count indicates the bit position in the quad-word ExAS header of the head of the backward path specification (explained momentarily). The turn pool is a container for turns. A turn pool logically contains two paths: a forward path and a backward path. The forward path indicates the path from a packets current fabric position to its destination. The backward path, via a transformation, indicates the path from a packets current fabric position back to the packet's origin. The bit count partitions the turn pool into its forward and backward components. The backward path of a packet when the packet is at its origin is null.

[0035] A turn is a variable sized, contiguous field containing one or more bits. The actual value of a turn is called its turn value. A turn value indicates the position of an output port relative to the input port on which a packet is received, and contains the number of relative ports to be skipped between the input port and the output port. An output port number is equal to ([input_port_number+turn_value+1] modulo [N2+1]). For example, the turn value for a packet entering a switch on port x and leaving on port x+1 is 0, and the turn value for a packet entering port x and leaving port x+4 is 3.

[0036] The switches in an ExAS fabric can be classified by the number of bits, of turn pool they consumed. Logically each switch has N2+1 ports, where N is the number of bits of turn pool consumed by the switch. For example, a 3-bit switch logically has 9 ports, even if physically it only had 7 ports. From a relative routing perspective, port 0 follows port N2, or a turn value of 0 in a packet that arrived on physical port N2 would indicate that the packet should exit the switch on physical port 0. A switch cannot use more physical ports than its turn pool consumption indicates. For example a 4-bit switch can have at most 17 physical ports. Conversely, a switch could consume 5-bits of turn pool but only contain say 6 physical ports, though profligate consumption of turn pool bits is strongly discouraged. The sizes of the turns within a turn pool is determined by the switches that process the turn pool as it traverses a fabric. FIG. 3 shows the routing portion of a unicast header.

[0037] ii. Path Initialization

[0038] The origin of the packet is responsible for constructing the header of the packet, including the path. The method for assignment of paths and other packet header values to a packet by the origin is implementation specific.

[0039] The turn pool of an as yet unrouted packet can be partitioned into two sections: the active partition and the unused partition. The active partition of a path contains the bits that are processed by the switches between the packet's origin and its destination. The unused portion of the turn pool contains any additional bits remaining in the turn pool that are not required to route the packet to its destination. For example, a packet that is expected to traverse six 4-bit switches on its route from origin to terminus would have a 24-bit active partition and a 16 bit unused partition. It is possible that one of partition may be empty. For example a ten 4-bit switch path would require all 40 bits of the turn pool and the unused partition would be empty. Conversely, two ExAS endpoints that were directly connected would not traverse any switches and the active partition of the turn pool would be empty.

[0040] When an origin constructs a path, it must supply two values: the turn pool and the bit count. When routing packets to endpoints the bit count is always initialized to be zero. When routing packets to switches, the bit count must be biased. For a packet to be accepted by a switch its turn count must be 23 when it arrives at the switch. To ensure this necessary condition, an endpoint that wishes to communicate with a switch must set the initial bit count of switch based packets to be the 23 plus the bit size of the active turn pool partition.

[0041] The active partition of the turn pool must be initially be right justified in the turn pool, with the unused partition filling the unused bits to the right of the active partition. The unused portion of an endpoint destined turn pool must set to all zeroes. The unused portion of an switch destined turn pool must set to all ones.

[0042] iii. Switch Path Transform

[0043] An N-bit switch, when it receives a path-routed packet, uses bits 25 through 24+N of the routing header (bits 0 through (N-1) of the turn pool) as the switches turn value. Before forwarding a path routed packet a switch must perform a switch path transform on the packet's path. A switch path transform consists of decrementing the packets bit count by N and shifting the turn pool N bits to the right, shifting in from the left the bit-reversed turn value used by the switch. For example, a 4-bit switch that received a packet with a turn pool value of 00 . . . 001001100100011b and a bit count of 63 would forward the packet with a turn pool value of 110000 . . . 00100110010b and a bit count of 59. If the next switch the aforementioned packet encountered consumed 8-bits of turn pool, the next transformed path would be 010011001100000 . . . 001b and the bit count would be 51.

[0044] The equations for switch based path transform are:

X=BIT_SIZE_OF_SWITCH

bit_count=bit_count−X

turn_pool=turn_pool[0:X−1], turn_pool[38:X]//concatenate

[0045] For example, the equations for a 4-bit switch path transform are:

bit_count=bit_count−4

turn_pool=turn_pool[0:3], turn_pool[38:4]//concatenate

[0046] FIG. 4 shows the mapping of a path routing turn pool field reversal.

[0047] The output port used by the origin of a packet is not a part of the packet's path, nor is the bit count decremented when it leaves its origin. The method for assignment of the output port of a packet at the origin is implementation specific.

[0048] iv. Endpoint Path Transform

[0049] A path from an origin to a terminus can be translated to provide a path from the terminus back to the origin. This translation, or path transform, requires that bit count be set to zero and the turn pool be inverted and bit reversed. For example, if a packet arrived at an endpoint with a turn pool of 011111111110000000000000000000000011110b, then the transformed path would be 100011111111111111111111111100000000001b band the bit count would be 0.

[0050] When a path transform is applied to the blind broadcast path identifier received during fabric enumeration, the result is a path to the root. This path to the root is used for event packets when the event tables are not programmed. Switches must independently record the input port in which a packet arrives if they wish to be able to perform a path transformation and return the packet, or trace its path back to its origin. The input port on a component is not captured within the path and must be maintained separately.

[0051] The terminus of a packet addressed by a transformed path as described above must be an endpoint.

[0052] b. Path Routing Errors

[0053] If a path is programmed incorrectly or if a port along the path is down, a path routed packet can encounter an error that prevents it from being forwarded any further. The following are path routed errors that must be detected by switches:

[0054] Bad Path: bit count=24 at a switch. If a packet is received that has a turn count of 24, and is NOT an Interface 1 packet, then the packet must be discarded as it is assumed that the path is bad. This prevents packets from circulating endlessly within the fabric. The switch drops the packet (crediting the packet in the process) and generates a Path Event packet to the origin of the packet.

[0055] Port Down. A packet cannot be forwarded because the output port is down. The switch drops the packet, and generates a Path Event to the origin.

[0056] Bad Turn Value: Port number is not mapped to a physical link at the node. Treated the same way as a Port Down described above.

[0057] When a path-routed error occurs, path events are generated for path routed packets. Path events are not generated for multicast packets. Multicast packets result in a multicast distribution error being sent to the multicast event host. A path event is sent back to the origin of a packet, and allows the origin to invalidate that path for use until it is repaired. Path events are discussed in more detail in Section

[0058] c. Multicast Routing

[0059] Multicast routing enables a single packet generated by an origin to be sent to multiple endpoints. Duplicate packets are generated only when the paths of the multicast packet diverge. From the ExAS perspective, multicast packets are restricted to writes; there is no such thing as a multicast read. A multicast packet is identified when a packet is received with a hop/bit count of 0-22.

[0060] Multicast packets are routed from node to node based on table look-ups at each switch. Each multicast packet has a Multicast Group Index, which identifies to multicast group to which the packet belongs. This 12-bit Multicast ID occupies bits 63-48 of the ExAS routing header. The bit count field of the unicast header is recast as a hop count field for multicast headers. A multicast distribution tree is limited to 22 switch traversals (hops). When a multicast packet is received at a switch, its hop count field (bits 24-19 of the ExAS routing header) is tested. If the hop count is 22, then a multicast routing error has occurred. If the hop count is 21 or less, then the hop count is incremented and the packet is forwarded.

[0061] Multicast routing is optional in ExAS. Multicast support is different for endpoints and switches. Endpoint support for multicast only requires the ability to inject multicast packets into a fabric. For a switch, multicast requires the implementation of a multicast routing table and multicast packet distribution mechanisms. Optional multicast support extends to the number of multicast groups supported by a switch. A switch's capability list structure includes a structure that specifies the number of multicast groups a switch supports. Endpoint capability lists include only a single bit that indicates whether multicast traffic is supported. For an endpoint, multicast support is all or nothing. Endpoints that support multicast must support all 4096 possible multicast groups.

[0062] ExAS supports a multicast routing model in which an endpoint can be a target of a multicast group a source of a multicast group, or both. That is, an endpoint can write into a multicast group to which it does not listen, listen to a group into which it does not write, or both write and listen. From the group perspective, a group can have a single writer or multiple writers and a single listener or multiple listeners.

[0063] Multicast distribution is switch port based. ExAS switches contain an array of multicast distribution vectors. A multicast distribution vector contains a pair of bits per switch port. These bits are the get and put bits. If the get bit for a given port is set, then that port is allowed to forward packets of the associated multicast group. If the get bit for a given port is not set, then receipt of a multicast packet on that port results in the packet being discarded and the generation is a multicast distribution error. When a valid multicast packet arrives at a switch, the put bits are used to determine on which ports the packet will exit. The port on which a packet arrives is excluded from the forwarding vector. If a packet arrives at a switch that has a null forwarding vector, then the packet is discarded and a multicast distribution error is generated.

[0064] A multicast manager configures multicast. A fabric's multicast manager is responsible for manipulating the multicast tables of a fabric's switches in a way that maintains a logically consistent view of multicast distribution.

[0065] A switch that receives a packet with a multicast group index larger than the switches set of supported multicast groups discards the packet and generates a multicast distribution error. A switch that does not support multicast is logically identical to a switch that supports 0 multicast groups.

[0066] There are no special routing requirements on terminal endpoints for multicast packets; they are received and processed as if they were unicast packets, with the caveat that path transform is not an option.

[0067] i. Generating Multicast Packets at an Endpoint

[0068] Generation of multicast packets at an endpoint is optional. The method for assignment of multicast groups and other header fields in an endpoint is implementation specific.

[0069] ii. Sending and Receiving Multicast Packets

[0070] When a switch receives a multicast packet, it must use the Multicast Group Index as an index into a Multicast Table to select a multicast group entry. The Multicast Table has an entry for each multicast group supported. The size of a multicast table entry is switch port size specific. The number of supported multicast groups is implementation specific, although a minimum recommended number for a switch is 32 (Multicast groups 0 through 31).

[0071] Each Multicast Table entry has the following format as depicted in FIG. 5, where N is the number of ports on the switch.

[0072] The multicast packets may be sent independently out each port. Credits and ordering are checked for each outgoing multicast packet independently of any other copies of the packet that may be sent out other ports. If a transmission error occurs when one of the multicast packets is sent, only that copy for the affected port is re-transmitted.

[0073] Note that a single multicast packet can result in the generation of many packets within the fabric. A switch may generate up to N2 multicast packets (where N is the bit number of the switch) for each multicast packet it receives. Applications need to be aware of this characteristic of multicast traffic and plan bandwidth usage accordingly.

[0074] iii. Multicast Routing Failures

[0075] There are several checks that a multicast packet must undergo in a switch before it can be forwarded, as summarized in Table 3. If a check fails, the multicast packet is discarded and an event is signaled. If multiple errors occur, the Failure Type used is listed from highest to lowest priority in the table. Multiple events may be set. Table 3 shows the different types of routing checks.

TABLE 3
Multicast Routing Checks
Multicast Check	Chip Event	Failure Type
Hop count = 22	S/W Multicast Distribution	Software Routing
	Failure	Failure
Unsupported Multicast	Unsupported Multicast	Software Routing
ID	Group ID	Failure
Input port mismatch	S/W Multicast Distribution	Software Routing
(protected groups only)	Failure	Failure
Output port(s) down	H/W Multicast Distribution	N/A
	Failure
No members in group	S/W Multicast Distribution	Software Routing
	Failure	Failure

[0076] The “Hop Count=22” failure means that the multicast write packet had a turn count set to the maximum value of 22 when entering a switch. Unlike a path-routed packet, a multicast packet that enters a switch with an erroneous bit/hop count does not cause a path event to be generated, however the multicast packet is discarded.

[0077] Unsupported Multicast ID means that the multicast write had a Multicast ID which does not have a corresponding entry in that switch's Multicast Table. For instance, a switch implements entries for multicast groups 0 through 31, but a multicast packet is received with a Multicast ID of 54.

[0078] If one or more of the output ports of a multicast group are down, the multicast packet is forwarded to the operational output ports that are group members.

[0079] 4. Protocol Encapsulation Interfaces

[0080] The PCI Express-AS specification several PEIs for ExAS architecture. They are detailed in the following sections.

[0081] 0 —Enumeration

[0082] 1 —Device Management

[0083] Multiple apertures

[0084] Partitioned by access rights

[0085] Fabric master, device host, bus driver, operating system

[0086] Specification defined registers vs. device specific registers

[0087] 2 —Event reporting

[0088] 3 -7—ExAS Fabric Management Reserved

[0089] 8 —PCI Express Core protocol

[0090] 9 —PCI Express MsgAS protocol

[0091] 10 —Simple Data Protocol

[0092] 11 -191—PCI-SIG defined interfaces

[0093] 192 -254—User defined interfaces

[0094] 255 —Reserved (Invalid)

[0095] a. Interface 0—Enumeration (Blind Broadcast)

[0096] Interface 0, blind broadcast, is a required interface that uses exactly one packet format, the blind broadcast packet. Any packet directed to interface 0 of a component will be interpreted as a blind broadcast packet. FIG. 6 shows the format of a blind broadcast packet and Table 4 describes it in further detail.

TABLE 4
Blind Broadcast Packet Field Descriptions
Bit	Bit field	Hop
Positions	width	Mutable	Field Name	Description
7:0	8	No	Protocol	This field identifies the packet as a blind broadcast
			Encapsulation	packet. All packets routed using PEI 0 are interpreted as
			Interface	blind broadcasts.
13:8	6	No	Credit Length	The credit length of a blind broadcast packet is 1. Any
				value other than 0 is a malformed packet.
18:14	5	No	Reserved
24:19	6	Yes	Bit Count	This field indicates the number of switch traversals the
				associated packet has undergone. Packets that are
				received by switches with a bit count values of 23 or less,
				or at s endpoints with a bit count value of 22 or less
				indicate that the receiving node has been pruned.
63:25	39	Yes	Turn Pool	The turn pool contains the variable bit width turn values
				of a path specification.
67:64	4	No	Operation	This field indicates the type of blind broadcast operation
				being performed. Operations are coded as follows:
				0 = Reset
				1 = Announce
				2 = Own
				3 = Re-Own
				4-15 = Reserved
71:68	4	No	PID Index	This field indicates upon which of a node's PIDs the
				associated operation is to be performed. A node that does
				not implement a packet's indicated PID silently discards
				the blind broadcast packet.
74:72	3	No	MTU	Accepted PID 0 blind broadcast packets set the receiving
				node's maximum transmission unit (MTU). Nominal
				MTU values are defined as follows:
				0 = 256
				1 = 512
				2 = 1024
				3 = 2048
				4 = 4096
				5-7 = Reserved
				Actual MTU values are the nominal value plus 64. The
				additional 64 bytes of MTU are a “fudge factor” added to
				the nominal value to account for the difference between
				an encapsulated packet's payload and the ExAS header
				and encapsulated packet's header overhead.
95:73	22	No	Implementation	The bits in this field can be used by implementations to
			Dependent	distribute instance specific data as part of a blind
			Fields	broadcast packet.
159:96	64	No	EUI	This field contains the NEUI of the origin of the blind
				broadcast packet.

[0097] In accordance with one embodiment of the present invention, there are 16 types of blind broadcast packets, only 4 of which are defined in revision 1 of the current ExAS specification. The remaining unspecified blind broadcast packet types are reserved in revision 1.

[0098] b. Interface 1—Node Configuration and Status

[0099] Interface 1, node configuration and status, is a required, encapsulation interface (it is a transport for another packet type) built using a restricted implementation of interface 6 (the simple data protocol (SDP) interface). Interface 0 SDP packet usage is limited to the 64-byte maximum data payload versions of the read and write packets. Interface 0 implements at least two pipes, pipe 0 and 1. Pipe 0 is the capability space pipe. It logically corresponds to the capability list structures of legacy PCI devices. Pipe 1 is the register space pipe. It logically corresponds to the registers of legacy PCI.

[0100] There are three packet types accepted by interface 1: a read packet a completion packet and a write packet. All three packet types are restricted to 64 dword packet payloads.

[0101] i. Interface 1 Read Request Packet

[0102] Simple Read Data Packet Format with Unicast Header Prepended for Purposes of Clarity is shown in FIG. 7 and further described in Table 5.

TABLE 5
SDP Read Packet Field Descriptions
Bit	Bit field
Positions	width	Field Name	Description
1:0	2	Type	This field indicates the type of the associated SDP packet. The
			possible types are:
			0 = Read Request
			1 = Read Completion
			2 = Write
			3 = Reserved
31:2	30	Offset	This field provides the offset of the desired data relative to the
			associated pipe index.
39:32	8	Reserved
46:40	7	Pipe Index	This field indicates the pipe index for the access. Interface 1
			supports 4 pipes (0-3) Pipe 0 contians the Configuration List.
			Pipe 1 contains the Register Space. Pipe 2 references an
			ancillary memory space. Pipe 3 references a read with side
			effects register space. Pipes 4-127 are reserved. Any interface 1
			packet that arrives with a pipe number larger than 3 results in a
			pipe access error.
47	1	CRC	This bit indicates whether the associated packet is protected by
			a CRC-32 value. When set, there is an additional dword of
			CRC appended to the packet. When clear, the packet ends
			immediately after the 8-byte header. There is no mechanism
			for requesting CRC on the response packet. The configuration
			of the read target will independently attach CRC to its
			completion packet, though implementations are free to use the
			use of CRC on a request to indicate that CRC should be
			appended to the completion.
53:48	6	Dwords	This field indicated the number of dwords, whether partial or
		Requested	complete being requested by the associated read request.
			Interface 1 data requests are always dword aligned. The skip
			and drop fields are used to provide sub-dword access. The
			dword field indicates the number of dwords more than one
			requested. There is no option for a null completion payload. A
			request must request at least one dword. That dword may be
			masked such that no valid data is returned, but even in the
			event of a completely masked dword being returned, a dword is
			returned. Thus a dword field value of 0 indicates that 1 dword
			is being requested and a dwords field containing the value 63
			indicates that 64 dwords are being requested.
55:54	2	Skip	This field indicates the number of bytes in the initial dword
			that are to be masked (disabled). The possible values indicate
			the following:
			0 = No bytes masked
			1 = Least significant byte of dword masked
			2 = Least significant 2 bytes of dword masked
			3 = Least significant 3 bytes of dword masked
			If a single dword is being requested, then the skip and drop
			fields reference the same dword. In this case, the logical “or”
			of the skip and drop masks are applied to the dword request.
57:56	2	Drop	This field indicates the number of bytes in the final dword that
			are to be masked (disabled). The possible values indicate the
			following:
			0 = No bytes masked
			1 = Most significant byte of dword masked
			2 = Most significant 2 bytes of dword masked
			3 = Most significant 3 bytes of dword masked
			If a single dword is being requested, then the skip and drop
			fields reference the same dword. In this case, the logical “or”
			of the skip and drop masks are applied to the dword request.
58	1	Request Type	Must be set. A packet format error will be generated if this bit
			is not set.
63:59	5	Transaction	This field indicated the source (requester) specific identifier for
		Number	the request. This field is used to match requests with responses.
			The source of a read request may apply values to this field in
			any manner deemed appropriate. It is the responsibility of the
			terminus of a read request to return the completion data in a
			packet with a matching transaction number.

[0103] ii. Interface 1 Completion Packets

[0104] Completions shall not be decomposed into decomposition completion sequences. An endpoint making, for example, a 64-dword request must be capable of receiving the entire response in a single 64-dword payload completion, and a node receiving a 64-dword request, must return all 64-dwords in a single packet. FIG. 8 depicts the format of an interface 1 completion data packet, which is more fully described in Table 6.

TABLE 6
SDP Completion Packet Field Descriptions
Bit	Bit field
Positions	width	Field Name	Description
1:0	2	Type	This field indicates the type of the associated SDP packet. The possible
			types are:
			0 = Read Request
			1 = Completion
			2 = Write
			3 = Reserved
7:2	6	Sequence	This field must be 0.
		Number
11:8	4	Completion	This field indicates the nature of the completion. When non-zero this field
		Type	indicates that an error was encountered that terminated the request
			prematurely. The encoding for this field are as follows:
			0 = Successful Completion
			1 = Time Out
			2 = Pipe Range Error
			3 = Pipe Protection Error
			4 = Access Error
			5-15 = Reserved
12	1	No Data	This bit indicates that there is no partial data being returned in association
			with a request failure. If this bit is set, then the Dwords field logically
			becomes −1. If the bit is not set and the completion type is non-zero, then
			the dwords field contains a number of dwords that is less than the number
			of dwords in the associated read request. After the first dword, which is
			covered by the skip field of the request, all subsequent dwords for a partial
			completion must be whole dwords. If a data request fails mid dword, then
			the dword in which the failure occurs is not returned with any other partial
			completion dwords.
14:13	2	Reserved
15	1	CRC	This bit indicates whether the associated packet is protected by a CRC-32
			value. When set, there is an additional dword of CRC appended to the
			packet. When clear, the packet ends immediately after the last indicated
			dword. The interface 1 header's dword size field does not include the CRC
			dword when present.
21:16	6	Payload	This field indicated the number of dwords, whether partial or complete
		Dwords	encompassed by the packet's payload. Interface 1 data requests are always
			dword aligned. Completions always return dword aligned payloads. If the
			request applied masks to the payload within the request, it is the duty of
			the requesting endpoint to “remember” the masks it supplied when
			requesting the data and to apply those masks again before using the data.
			The dword field indicates the number of dwords more than one requested.
			There is no option for a null completion payload. A request must request
			at least one dword. That dword may have been masked such that no valid
			data exists in the single dword payload returned. A dword field value of 0
			indicates the presence of a 1-dword payload and a dwords field containing
			the value 63 indicates the presence of a 64-dword payload.
25:22	4	Reserved
26	1	End	Must be set. A packet format error will be generated if this bit is not set.
31:27	5	Transaction	This field indicated the source (requester) specific identifier for the
		Number	request. The transaction number of a request must exactly match the
			transaction number present in the associated request packet.

[0105] iii. Interface 1 Write Packets

[0106] FIG. 9 shows the format of an interface 1 write data packet, which is more fully described in Table 7.

TABLE 7
SDP Write Packet Field Descriptions
Bit	Bit field
Positions	width	Field Name	Description
1:0	2	Type	This field indicates the type of the associated SDP packet. The possible
			types are:
			0 = Read Request
			1 = Read Completion
			2 = Write
			3 = Reserved
31:2	30	Offset	This field provides the offset of the associated data payload relative to the
			associated pipe index.
39:32	8	Reserved
46:40	7	Pipe Index	This field indicates the pipe index for the access. Interface 1 supports 4
			pipes (0-3) Pipe 0 contains the Configuration List. Pipe 1 contains the
			Register Space. Pipe 2 references an ancillary memory space. Pipe 3
			references a read with side effects register space. Pipes 4-127 are reserved.
			Any interface 1 packet that arrives with a pipe number larger than 3 results
			in a pipe access error.
47	1	CRC	This bit indicates whether the associated packet is protected by a CRC-32
			value. When set, there is an additional dword of CRC appended to the
			packet's payload. When clear, the encapsulated packet ends immediately
			after the payload.
53:48	6	Payload	This field indicated the number of dwords, whether partial or complete
		Dwords	being transferred. Interface 1 data writes are always dword aligned. The
			skip and drop fields are used to provide sub-dword access. The dword
			field indicates the number of dwords more than one being written. There is
			no option for a null write payload. A write must supply at least one dword
			of payload. That dword may be masked such that no valid data is being
			presented, but even in the event of a completely masked dword being
			written, a dword is present. Thus a dword field value of 0 indicates that 1
			dword is being written and a dwords field containing the value 63
			indicates that 64 dwords are being written.
55:54	2	Skip	This field indicates the number of bytes in the initial dword that are to be
			masked (disabled). The possible values indicate the following:
			0 = No bytes masked
			1 = Least significant byte of dword masked
			2 = Least significant 2 bytes of dword masked
			3 = Least significant 3 bytes of dword masked
			If a single dword is being written, then the skip and drop fields reference
			the same dword. In this case, the logical “or” of the skip and drop masks
			are applied to the dword being written.
57:56	2	Drop	This field indicates the number of bytes in the final dword that are to be
			masked (disabled). The possible values indicate the following:
			0 = No bytes masked
			1 = Most significant byte of dword masked
			2 = Most significant 2 bytes of dword masked
			3 = Most significant 3 bytes of dword masked
			If a single dword is being written, then the skip and drop fields reference
			the same dword. In this case, the logical “or” of the skip and drop masks
			are applied to the dword being written.
58	1	Write Type	Must be set. A packet format error will be generated if a type 2 packet is
			received with this bit is not set.
63:59	5	Transaction	Must be zero. A packet format error will be generated if a type 2 packet is
		Number	received with this field set to a non-zero value.

[0107] c. Interface 2—Event Notification

[0108] Interface 2, event notification, is an optional interface that uses exactly one packet format, the event notification packet. Any packet directed to interface 2 of a component will be interpreted as an event notification packet. For instance, a routing error packet format is shown in FIG. 10.

[0109] d. Interface 8—PCI-Express Core (ExC)

[0110] Interface 8, ExC, is an optional encapsulation interface targeting the extension of the ExC packet structure to ExAS. Both unicast and multicast ExAS packet formats may be used to tunnel ExC packets. The mechanism used to create an ExAS header for an ExC packet is implementation dependent. The ExC packet encodings for ExAS packets are ignored by ExAS proper (as is the content of all packets formats/protocols tunneled through ExAS), though it is reasonable to expect that the implementation specific ExAS header creation logic may have a significant interest therein.

[0111] e. Interface 10—Simple Data Protocol (SDP)

[0112] Interface 10, SDP, is an optional encapsulation interface targeting the extension of a simple memory like data movement protocol to ExAS. Both unicast and multicast ExAS packet formats may be used to tunnel SDP packets. SDP is a much simpler protocol than PCI-Express Core, and it targeted at simple partitioned memory space access between peer ExAS nodes.

[0113] i. SDP Pipes

[0114] SDP is pipe based. A pipe is a memory aperture. A pipe has an associated base address and size. The nature of the base address and size are implementation specific. A pipe also has a path protection capability. Each pipe has it own independent pair of path protection registers. Neither, one, or both of these path protection registers may be enabled. If neither of the path protection registers are enabled, then the pipe allows promiscuous access to its memory aperture. If one path protection register is enabled, then only packets with the matching path are allowed to access the associated aperture. If both path protection registers are enabled, then packets from either path are provided access. If path protection is enabled, then any packet arriving from a non-matching fabric location is rejected and a pipe protection event is generated.

[0115] ii. SDP Packet Formats

[0116] A. SDP Read Request Packets

[0117] FIG. 11 show and example of a simple read data packet format with ExAS Routing Header Attached shown for purposes of clarity. FIG. 12 shows a simple read data packet format. The simple read data packet format is further described in Table 8.

TABLE 8
SDP Read Packet Field Descriptions
Bit	Bit field
Positions	width	Field Name	Description
1:0	2	Type	This field indicates the type of the associated SDP packet. The
			possible types are:
			0 = Read Request
			1 = Read Completion
			2 = Write
			3 = Reserved
39:2	38	Offset	This field provides the offset of the desired data relative to the
			associated pipe index.
46:40	7	Pipe Index	This field indicates the pipe index for the access.
47	1	CRC	This bit indicates whether the associated packet is protected by
			a CRC-32 value. When set, there is an additional dword of
			CRC appended to the packet. When clear, the packet ends
			immediately after the 8-byte header. There is no mechanism
			for requesting CRC on the response packet. The configuration
			of the read target will independently attach CRC to its
			completion packet, though implementations are free to use the
			use of CRC on a request to indicate that CRC should be
			appended to the completion.
53:48	6	Dwords	This field indicated the number of dwords, whether partial or
		Requested	complete being requested by the associated read request.
			Interface 1 data requests are always dword aligned. The skip
			and drop fields are used to provide sub-dword access. The
			dword field indicates the number of dwords more than one
			requested. There is no option for a null completion payload. A
			request must request at least one dword. That dword may be
			masked such that no valid data is returned, but even in the
			event of a completely masked dword being returned, a dword is
			returned. Thus a dword field value of 0 indicates that 1 dword
			is being requested and a dwords field containing the value 63
			indicates that 64 dwords are being requested.
55:54	2	Skip	This field indicates the number of bytes in the initial dword
			that are to be masked (disabled). The possible values indicate
			the following:
			0 = No bytes masked
			1 = Least significant byte of dword masked
			2 = Least significant 2 bytes of dword masked
			3 = Least significant 3 bytes of dword masked
			If a single dword is being requested, then the skip and drop
			fields reference the same dword. In this case, the logical “or”
			of the skip and drop masks are applied to the dword request.
57:56	2	Drop	This field indicates the number of bytes in the final dword that
			are to be masked (disabled). The possible values indicate the
			following:
			0 = No bytes masked
			1 = Most significant byte of dword masked
			2 = Most significant 2 bytes of dword masked
			3 = Most significant 3 bytes of dword masked
			If a single dword is being requested, then the skip and drop
			fields reference the same dword. In this case, the logical “or”
			of the skip and drop masks are applied to the dword request.
58	1	Request Type	This bit, when set, indicates whether the associated packet's
			bits 57:48 contain a 2-bit skip, 2-bit drop and 6-bit dwords
			requested fields or a 10 bit octword (16-byte) aligned octwords
			requested field.
63:59	5	Transaction	This field indicated the source (requester) specific identifier for
		Number	the request. This field is used to match requests with responses.
			The source of a read request may apply values to this field in
			any manner deemed appropriate. It is the responsibility of the
			terminus of a read request to return the completion data in a
			packet with a matching transaction number.

[0118] B. SDP Completion Packets

[0119] FIG. 13 shows a simple completion data packet format, which is further described in Table 9.

TABLE 9
SDP Read Completion Packet Field Descriptions
Bit	Bit field
Positions	width	Field Name	Description
1:0	2	Type	This field indicates the type of the associated SDP packet. The possible
			types are:
			0 = Read Request
			1 = Completion
			2 = Write
			3 = Reserved
7:2	6	Decomposition	A request that is larger than a single fabric MTU needs to be decomposed
		Sequence	into a series of completion frames. This field indicates the absolute
		Number	decomposition sequence number of a packet. A packet with a
			decomposition sequence number of 0 and its end bit set indicates that the
			packet contains all the requested data, or that an error occurred and no
			additional data will be forth coming.
11:8	4	Completion	This field indicates the nature of the completion. When non-zero this
		Type	field indicates that an error was encountered that terminated the request
			prematurely. The encoding for this field are as follows:
			0 = Successful Completion
			1 = Time Out
			2 = Pipe Range Error
			3 = Pipe Protection Error
			4 = Access Error
			5-15 = Reserved
			If the completion type is non-zero then the packet's end bit must also be
			set.
12	1	No Data	This bit indicates that there is no partial data being returned in
			association with a request failure. If this bit is set, then the Dwords field
			logically becomes −1. If the bit is not set and the completion type is non-
			zero, then the dwords field contains a number of dwords that is less than
			the number of dwords in the associated read request. After the first
			dword, which is covered by the skip field of the request, all subsequent
			dwords for a partial completion must be whole dwords. If a data request
			fails mid dword, then the dword in which the failure occurs is not
			returned with any other partial completion dwords.
14:13	2	Reserved
15	1	CRC	This bit indicates whether the associated packet is protected by a CRC-32
			value. When set, there is an additional dword of CRC appended to the
			packet. When clear, the packet ends immediately after the last indicated
			dword. Interface 10 header's dword size field does not include the CRC
			dword when present.
25:16	10	Payload Dwords	This field indicated the number of dwords, whether partial or complete
			encompassed by the packet's payload. Interface 1 data requests are
			always dword aligned. Completions always return dword-aligned
			payloads. If the request applied masks to the payload within the request,
			it is the duty of the requesting endpoint to “remember” the masks it
			supplied when requesting the data and to apply those masks again before
			using the data. The dword field indicates the number of dwords more
			than one requested. There is no option for a null completion payload. A
			request must request at least one dword. That dword may have been
			masked such that no valid data exists in the single dword payload
			returned. A dword field value of 0 indicates the presence of a 1-dword
			payload and a dwords field containing the value 63 indicates the presence
			of a 64-dword payload.
26	1	End	This bit, when set, indicates that this is the final packet in a
			decomposition sequence. That is, that the request has been fully satisfied,
			or has terminated in an error.
31:27	5	Transaction	This field indicated the source (requester) specific identifier for the
		Number	request. The transaction number of a request must exactly match the
			transaction number present in the associated request packet.

[0120] C. SDP Write Packets

[0121] FIGS. 14 and 15 show exemplary simple write data packet formats, further described in Table 10.

TABLE 10
SDP Write Packet Field Descriptions
Bit	Bit field
Positions	width	Field Name	Description
1:0	2	Type	This field indicates the type of the associated SDP packet. The possible
			types are:
			0 = Read Request
			1 = Read Completion
			2 = Write
			3 = Sequenced Write
39:2	38	Offset	This field provides the offset of the associated data payload relative to
			the associated pipe index.
46:40	7	Pipe Index	This field indicates the pipe index for the access.
47	1	CRC	This bit indicates whether the associated packet is protected by a CRC-32
			value. When set, there is an additional dword of CRC appended to the
			packet's payload. When clear, the encapsulated packet ends immediately
			after the payload.
57:48	10	Payload Dwords	If write type = 0, then this field indicates the number of dword aligned
or	or		dwords in the packet's payload. All bytes in the payload are valid.
53:48	6		If Write Type = 1, then this field indicates the number of dwords,
			whether partial or complete being transferred. Interface 1 data writes are
			always dword aligned. The skip and drop fields are used to provide sub-
			dword access. The dword field indicates the number of dwords more than
			one being written. There is no option for a null write payload. A write
			must supply at least one dword of payload. That dword may be masked
			such that no valid data is being presented, but even in the event of a
			completely masked dword being written, a dword is present. Thus a
			dword field value of 0 indicates that 1 dword is being written and a
			dwords field containing the value 63 indicates that 64 dwords are being
			written.
55:54	2	Skip	This field indicates the number of bytes in the initial dword that are to be
			masked (disabled). The possible values indicate the following:
			0 = No bytes masked
			1 = Least significant byte of dword masked
			2 = Least significant 2 bytes of dword masked
			3 = Least significant 3 bytes of dword masked
			If a single dword is being written, then the skip and drop fields reference
			the same dword. In this case, the logical “or” of the skip and drop masks
			are applied to the dword being written.
57:56	2	Drop	This field indicates the number of bytes in the final dword that are to be
			masked (disabled). The possible values indicate the following:
			0 = No bytes masked
			1 = Most significant byte of dword masked
			2 = Most significant 2 bytes of dword masked
			3 = Most significant 3 bytes of dword masked
			If a single dword is being written, then the skip and drop fields reference
			the same dword. In this case, the logical “or” of the skip and drop masks
			are applied to the dword being written.
58	1	Write Type	The write type field indicates whether bits 57:48 of the write header
			contains a 6 bit dword count with skip and drop masks (bit is set), or a 10
			bit, dword aligned size (bit is clear).
63:59	5	Transaction	Transaction number 0 indicates that the write packet is an isolated
		Number	transaction. That is, it is not part of a sequenced write.
			A non-zero transaction number indicates that this packet is the first in a
			series of write transactions. All subsequent writes in the series will share
			this packet's transaction number.

[0122] D. SDP Sequenced Write Packets

[0123] FIGS. 16 and 17 show exemplary simple sequenced write data packet formats, described in more detail in Table 11.

TABLE 11
SDP Sequenced Write Packet Field Descriptions
Bit	Bit field
Positions	width	Field Name	Description
1:0	2	Type	This field indicates the type of the associated SDP packet. The possible
			types are:
			0 = Read Request
			1 = Read Completion
			2 = Write
			3 = Sequenced Write
7:2	38	Decomposition	This field indicates the relative position in a decomposition sequence of
		Sequence	the associated packet. For a given sequenced write, all sequence numbers
		Number	must be strictly sequential with the exception of sequence numbers
			wrapping from sequence number 63 to 0. Sequence numbers wrap and
			hence indeterminately large write sequences are possible. Following
			initialization of a write sequence by a write packet with a non-zero
			transaction number, the decomposition sequence number of the initial
			sequenced write packet is 0.
14:8	7	Pipe Index	This field indicates the pipe index for the access.
15	1	CRC	This bit indicates whether the associated packet is protected by a CRC-32
			value. When set, there is an additional dword of CRC appended to the
			packet's payload. When clear, the encapsulated packet ends immediately
			after the payload.
56:48	9	Payload	If sequenced write type = 0, then this field indicates the number of
		Quadwords	quadword aligned quadwords in the packet's payload. All bytes in the
			payload are valid.
57	1	End	If sequenced write type = 0, then this bit, when set, indicates that this is
			the last write in the sequence, or when clear, indicates that additional
			packets for this sequence are to follow.
55:48	8	Payload Dwords	If sequenced write type = 1, then this field indicates the number of dword
			aligned bytes in the payload. The least drop field indicates whether the
			final dword of the sequenced write is a partial dword.
57:56	2	Drop	If sequenced write type = 1, then this field indicates the number of bytes
			in the final dword that are to be masked (disabled). The possible values
			indicate the following:
			0 = No bytes masked
			1 = Most significant byte of dword masked
			2 = Most significant 2 bytes of dword masked
			3 = Most significant 3 bytes of dword masked
			If a single dword is being written, then the skip and drop fields reference
			the same dword. In this case, the logical “or” of the skip and drop masks
			are applied to the dword being written.
58	1	Sequenced Write	The sequenced write type field indicates whether bits 57:48 of the
		Type	sequenced write header contains a 2-bit drop field and an 8-bit dword
			count field (bit is set), or a termination bit and a 9-bit quadword count
			(bit is clear). If the sequenced write type bit is set, then the associated
			packet is the final packet of the sequenced write sequence.
63:59	5	Transaction	Transaction number 0 indicates that the write packet is an isolated
		Number	transaction. That is, it is not part of a sequenced write.
			A non-zero transaction number indicates that this packet is the first in a
			series of write transactions. All subsequent writes in the series will share
			this packet's transaction number.

[0124] Having now described preferred embodiments of the invention, it should be apparent to those skilled in the art that the foregoing is illustrative only and not limiting, having been presented by way of example only. All the features disclosed in this specification (including any accompanying claims, abstract, and drawings) may be replaced by alternative features serving the same purpose, and equivalents or similar purpose, unless expressly stated otherwise. Therefore, numerous other embodiments of the modifications thereof are contemplated as falling within the scope of the present invention as defined by the appended claims and equivalents thereto.

[0125] For example, the techniques may be implemented in hardware or software, or a combination of the two. Preferably, the techniques are implemented in computer programs executing on programmable computers that each include a processor, a storage medium readable by the processor (including volatile and non-volatile memory and/or storage elements), at least one input device and one or more output devices. Program code is applied to data entered using the input device to perform the functions described and to generate output information. The output information is applied to one or more output devices.

[0126] Each program is preferably implemented in a high level procedural or object oriented programming language to communicate with a computer system, however, the programs can be implemented in assembly or machine language, if desired. In any case, the language may be a compiled or interpreted language.

[0127] Each such computer program is preferably stored on a storage medium or device (e.g., CD-ROM, RAM, ROM, hard disk or magnetic diskette) that is readable by a general or special purpose programmable computer for configuring and operating the computer when the storage medium or device is read by the computer to perform the procedures described in this document. The system may also be considered to be implemented as a computer-readable storage medium, configured with a computer program, where the storage medium so configured causes a computer to operate in a specific and predefined manner. For illustrative purposes the present invention is embodied in the system configuration, method of operation and product or computer-readable medium, such as floppy disks, conventional hard disks, CD-ROMS, Flash ROMS, nonvolatile ROM, RAM and any other equivalent computer memory device. It will be appreciated that the system, method of operation and product may vary as to the details of its configuration and operation without departing from the basic concepts disclosed herein.

[0128] In the manner described above, the present invention thus provides a system and method to select a fabric master. While this invention has been described with reference to the preferred embodiments, these are illustrative only and not limiting, having been presented by way of example. Other modifications will become apparent to those skilled in the art by study of the specification and drawings. It is thus intended that the following appended claims include such modifications as fall within the spirit and scope of the present invention.

Claims

1. A switch fabric network for creating a packet encapsulating a protocol comprising: a protocol encapsulation interface identifier; packet field data; a packet header using said protocol encapsulation interface identifier and packet field data; and a first node adapted to create said packet header and to transmit said packet header to a second node.

2. A system of claim 1 wherein said packet field data is comprised of a bit count and a turn pool;

3. A system of claim 1 wherein said packet field data is comprised of a hop count and a multisource index.

4. A system of claim 1 wherein said packet field data is comprised of a credit length, a bit count, a turn pool, an operation, a PID index, an MTU, and an EUI.

5. A system in a switch fabric network for creating a packet encapsulating a protocol comprising: a packet type identifier; packet field data; a packet header using said packet type identifier and packet field data; and a first node adapted to create said packet header and to transmit said packet header to a second node.

6. A system according to claim 5 wherein said packet field data is comprised of an offset, a pipe index, a CRC, a field indicating dwords requested, a skip identifier; a drop identifier, a request type, and a transaction number.

7. A system according to claim 5 wherein said packet field is comprised of a sequence number, a completion type, a CRC, a payload dwords, an end identifier, and a transaction number.

8. A system according to claim 5 wherein said packet field is comprised of an offset, a pipe index, a CRC, a payload dwords, a skip identifier; a drop identifier; a write type, and a transaction number.

9. A system according to claim 5 wherein said packet field is comprised of a decomposition sequence number, a completion type, a CRC, payload dwords, an end identifier, and a transaction number.

10. A system according to claim 5 wherein said packet field is comprised of a decomposition sequence number, a pipe index, a CRC, a first payload dwords, an end identifier, a second payload dwords, a drop identifier, a sequenced write type, and a transaction number.

11. A computer-readable medium having stored thereon a data structure representing a packet comprising: a first field comprising data representing a protocol encapsulation interface; a field set comprising a second field comprising data representing a credit length; a third field comprising data representing a traffic class; a fourth field representing a bypass queue; a fifth field representing a credit type; a sixth field representing a bit count; and a seventh field representing a turn pool.

12. A computer-readable medium having stored thereon a data structure comprising: a first field comprising data representing a protocol encapsulation interface; a second field comprising data representing a credit length; a third field comprising data representing a traffic class; a fourth field representing a hop count; a fifth field representing a multicast source index; a sixth field representing a bit count; and a seventh field representing a turn pool.