Abstract: A computer system comprises a plurality of shelves. Each shelf has a carrier for removably receiving a plurality of information processing modules and a switching module. Each shelf also has an interconnection member for providing connections between the information processing modules and the switching module. The switching modules of the respective shelves are interconnected in a logical stacking configuration to form a logical stacking arrangement.

Abstract: A computer system comprises a plurality of shelves. Each shelf has a carrier for removably receiving a plurality of information processing modules and a switching module. Each shelf also has an interconnection member for providing connections between the information processing modules and the switching module. The shelves are logically connected into a plurality of stacks, the switching modules of the respective shelves in each stack being interconnected in a logical stacking configuration.

Abstract: A method and apparatus for providing spanning tree support are provided. According to one aspect of the present invention, a network device includes two or more ports that are part of a trunk. One of the two or more ports are selected for participation in a loop-free topology discovery protocol. Then, the loop-free topology discovery protocol is executed for the selected port. If the loop-free topology discovery protocol indicates the selected port is to be blocked, then all of the ports of the trunk are blocked. According to another aspect of the present invention, a set of states for association with each port of a network device is provided. The set of states includes a “blocked” state in which both learning and forwarding are inhibited, a “learn only” state in which learning is permitted and forwarding remains inhibited, and a “non-blocked” state in which both learning and forwarding are permitted.

Abstract: A central node in a network computes for, and sends to, each node a forwarding table which consists of the set of neighbors to which the node should forward a message intended for a particular destination. The message includes a version number in the packet field header indicating which forwarding table version the node should use to forward the packet. The node does not begin marking and forwarding packets according to the new version number immediately. The node may wait a period of time after receiving the new table or may wait until receiving notification from the fabric manager to begin using the new version number. When a node receives a message from an end node, it inserts either the most recently received version number in one embodiment or uses the version dictated by the fabric manager in another embodiment.

Abstract: A system and method for sharing a resource (e.g., buffers) among multiple queues (e.g., InfiniBand send queues). Each queue has a local context indicating the Current amount of the resource allocated to the queue, the Minimum it is guaranteed, and a Maximum amount. A context may also include one or more pointers for managing the queue. Global Available and Reserve values indicate, respectively, the amount of the resource not currently allocated and the amount of the resource being reserved to guarantee each queue its Minimum. When an element is posted to a queue, a Full signal is raised for every queue to prevent further activity while the global values and the affected queue's context are updated. Then, if Available>0, the signal is de-asserted for any queue for which Current<Minimum or, if Available>Reserve, for any queue for which Current<Maximum.

Abstract: A system and method for scheduling communications from a communication interface. Each of multiple send queues is associated with a destination and assigned to a logical communication channel. A list of stalled queues identifies those send queues for which a constraint restricts scheduling. When a queue is to be selected for service, a scheduler first attempts to find a member of the stalled queues list that is no longer stalled (e.g., any constraints were satisfied). Such a queue is selected for service if it exists. If there is no such queue, then all logical channels are examined and, from a list of send queues assigned to a selected logical channel, a send queue is selected for servicing. After a queue is scheduled from the stalled queues list, the list of queues assigned to its logical channel is adjusted (e.g., to place the queue at the tail of the list).

Abstract: An apparatus and method for managing the receipt of communication traffic in the form of packets or other units. The apparatus includes a communication interface (e.g., a NIC, a TCA) coupled to one or more host computer systems. Through Direct Memory Access (DMA) operations, the interface reassembles payloads of received packets into host buffers based on their sequence numbers, without buffering them in the interface. Packet headers are separated from the payloads and passed to a host for protocol processing after the payload DMA is completed. Host buffers may be of virtually any size. For each communication connection, state information is maintained on the interface, which may identify an upper level protocol so that an upper level protocol header is passed to the host as part of the packet header, not as part of the payload. Protocol termination remains in the host.

Abstract: An apparatus is described comprising: a switch for providing a plurality of communication channels between a plurality of nodes; and a crossbar switch communicatively coupled between the switch and the nodes for allocating one or more of a plurality of links to each of the nodes. Additionally, in a system including a switch for providing a plurality of communication channels between a plurality of nodes, a method is disclosed comprising the steps of: determining bandwidth requirements of each node in the system; and allocating links to the nodes based on the bandwidth requirements.

Abstract: A method and apparatus for interconnecting a first device and a second device in a network. The first device and the second device are connected to a plurality of interfaces. The plurality of interfaces emulate a single high-speed interface. According to an embodiment of the present invention, a first identifier is assigned to the first interface and the second interface at the first device. According to another embodiment of the present invention, one of the plurality of interfaces is selected to transmit a packet of data.

Abstract: A modular computer system may be provided. The modular computer system may comprise a carrier operable removably to receive a plurality of computer system modules therein. A plurality of information processing modules can be removably received in the carrier, each module may have a communications port operable to connect to a communications network internal to the carrier. The modular computer system may also comprise a switch operable to connect to the internal communications network to distribute information messages between the modules and to connect to an external communications network. An information distribution module may be provided removably received in the carrier operable connect to the internal communications network to receive an information message, to perform processing on the message to determine a destination, and to forward the message toward the determined destination via the internal communications network.

Abstract: A protocol processor is a specialized processor dedicated to extracting data (i.e., a payload) from an incoming communication (e.g., packet) and configuring outgoing data for transmission, and includes one or more protocol processing elements (PPE). Each PPE may be programmed to handle any set of communication protocols, and includes a set of large registers (e.g., 128 bytes, 256 bytes). A PPE also includes a parse unit for parsing packets to retrieve certain information and a lookup unit for accessing a control block indicating how to process an incoming packet or outgoing payload. A modification unit removes headers from incoming packets and/or adds them to outgoing data. A timer unit manages a large number of timers (e.g., for different communication streams). A control block cache may store recently accessed control blocks, and a data streaming unit streams packets (or packet portions) into and out of the registers.

Abstract: An architecture for a highly integrated network element building block is provided. According to one aspect of the present invention, a network device building block includes a network interface with multiple ports for transmitting and receiving packets over a network. The network device building block also includes a packet buffer storage which is coupled to the network interface. The packet buffer storage acts as an elasticity buffer for adapting between incoming and outgoing bandwidth requirements. A shared memory manager may also be provided dynamically allocate and deallocate buffers in the packet buffer storage on behalf of the network interface and other clients of the packet buffer storage. The network device building block further includes a switch fabric which is coupled to the network interface. The switch fabric provides forwarding decisions for received packets. A given forwarding decision includes a list of ports upon which a particular received packet is to be forwarded.

Abstract: A system and method for updating packet headers using hardware that maintains the high performance of the network element. In one embodiment, the system includes an input port process (IPP) that buffers the input packet received and forwards header information to the search engine. The search engine searches a database maintained on the switch element to determine the type of the packet. In one embodiment, the type may indicate whether the packet can be routed in hardware. In another embodiment, the type may indicate whether the packet supports VLANs. The search engine sends the packet type information to the IPP along with the destination address (DA) to be updated if the packet is to be routed, or a VLAN tag if the packet has been identified to be forwarded to a particular VLAN. The IPP, during transmission of the packet to a packet memory selectively replaces the corresponding fields, e.g., DA field or VLAN tag field; the modified packet is stored in the packet memory.

Abstract: A distributed multi-layer network element delivering Layer 2 (data link layer) wire-speed performance within and across subnetworks, allowing queuing decisions to be based on Layer 3 (network layer) protocol and endstation information combined with Layer 2 topology information. The network element performs packet relay functions using multiple switching subsystems as building blocks coupled to each other to form a larger switch that acts as both a router and a bridge. Each switching subsystem includes a hardware forwarding search engine having a switching element coupled to a forwarding memory and an associated memory. The switching subsystems and their fully meshed interconnection allow the network element to scale easily without compromising packet forwarding speed and without significantly increasing the storage requirements of each forwarding memory.

Abstract: A multi-layer network element for forwarding received packets from an input port to one or more output ports. The packet is examined to look for first and second forwarding information. A packet is also assigned to a class and provided with default packet forwarding information. An associative memory is searched once for each type of information. The results from the two searches are combined with the default packet forwarding information to forward the packet to the appropriate one or more output ports. In some instances, the results of the first search dominate the forwarding decision, in other, the results of the second search dominate the forwarding decision, and in still other instances, the default information dominates.

Abstract: A method and apparatus for providing spanning tree support are provided. According to one aspect of the present invention, a network device includes two or more ports that are part of a trunk. One of the two or more ports are selected for participation in a loop-free topology discovery protocol. Then, the loop-free topology discovery protocol is executed for the selected port. If the loop-free topology discovery protocol indicates the selected port is to be blocked, then all of the ports of the trunk are blocked. According to another aspect of the present invention, a set of states for association with each port of a network device is provided. The set of states includes a "blocked" state in which both learning and forwarding are inhibited, a "learn only" state in which learning is permitted and forwarding remains inhibited, and a "non-blocked" state in which both learning and forwarding are permitted.

Abstract: A multi-layer network element for forwarding received packets from an input port to one or more output ports. The packet is examined to look for different types of forwarding information. An associative memory is searched once for each type of information. The results from the two searches are combined to forward the packet to the appropriate one or more output ports. The packet may be examined for other information as well to make the forwarding decisions. In one embodiment, the invention examines the packet for layer 2 information as the first type and layer 3, and perhaps some layer 4, information as the second type. The results are merged to determine the most appropriate combination of layer 2 or layer 3 forwarding decisions for the packet.

Abstract: A network system dynamically controls data flow between physical links by logically combining multiple physical links into a single logical channel trunk, preferably to balance data flow carried by each link. Each link in the trunk has identical physical layer and identical media access control layer characteristics. A system server assigns a single media access control layer address to the single trunked logical channel, preferably randomly by hashing destination media access control layer addresses for the links. The system server includes, in addition to a physical layer and a network layer, a pseudo-driver software layer disposed therebetween, which pseudo-driver software layer functions as a multiplexer in a receive path and functions as a de-multiplexer in a transmit path. The resultant preferably Ethernet-compatible network system operates in full-duplex mode and distributes packets from the server to the links to preserve temporal order of data flow.

Abstract: A method and apparatus for shared memory management in a switched network element is provided. According to one aspect of the present invention, a shared memory manager for a packet forwarding device includes a pointer memory having stored therein information regarding buffer usage (e.g., usage counts) for each of a number of buffers in a shared memory. An encoder is coupled to the pointer memory for generating an output which indicates a set of buffers that contains a free buffer. The shared memory manager further includes a pointer generator that is coupled to the encoder for locating a free buffer in the set of buffers. The pointer generator is further configured to produce a pointer to the free buffer based upon the output of the encoder and the free buffer's location within the set of buffers.

Abstract: A method and apparatus for providing trunking support in a network device is provided. According to one aspect of the present invention, a network device includes at least one port that is configured to be included in a trunk. The network device also includes a memory for storing a forwarding database. The forwarding database includes entries containing therein forwarding information for a subset of network addresses. The network device further includes a learning circuit. The learning circuit is coupled to the trunked port and the memory. The learning circuit is configured to modify the forwarding database to reflect an association between the trunked port and a first address contained within a packet received by the trunked port. If the trunk is of a first type, the learning circuit updates the forwarding database based upon a trunk designator corresponding to the trunk; otherwise, the learning circuit updates the forwarding database based upon a port designator corresponding to the trunked port.