Abstract: System and method for using multiple global identification subnet prefix values in a network switch environment in a high performance computing environment. A packet is received from a network fabric by a first Host Channel Adapter (HCA). The packet has a header portion including a destination subnet prefix identifying a destination subnet of the network fabric. The network HCA is allowed to receive the first packet from a port of the network HCA by selectively determining a logical state of a flag and, selectively in accordance with a predetermined logical state of the flag, ignoring the destination subnet prefix identifying the destination subnet of the network fabric.

Abstract: In an apparatus and method for providing device sharing, a first plurality of upstream ports are each connectable to a respective host and at least one downstream port is connectable to a device. A virtual port is defined that is associated a routing table to effect device virtualization by redirection of information packets received by the virtual port.

Abstract: Additional routing information is added to a transaction packet without recalculating an ECRC by inserting the additional routing information at predetermined bit positions in the transaction packet known to have predetermined bit values. The modified transaction packet can then be routed using the additional routing information. The predetermined bit values can subsequently be reinstated at the predetermined bit positions in the packet.

Abstract: A method is provided for supporting device sharing between hosts via a bus fabric. A master host owns a device tree and provides IO services to at least one client host. The client host comprises generic device drivers and subscribes to IO services provided by the master host.

Abstract: An interconnect apparatus, for example a switch, supports PCI-Express. The apparatus has a first plurality of ports configurable as upstream ports, each connectable to a respective host, and at least one port configurable as a downstream port connectable to a device. The apparatus is operable to support sharing of the device resources between hosts.

Abstract: At an ingress point to a shared transaction infrastructure, for example a shared PCI Express infrastructure, an entry in a segment table maps an address in a transaction packet to a target for the transaction packet. The entry in the segment table identifies one of a plurality of channel descriptors providing one or more of forwarding information for the transaction packet and information for constructing an additional header for the transaction packet.

Abstract: PCI Express transactions can be transmitted via a shared PCI Express infrastructure. At an infrastructure ingress point an additional header comprising at least a source identifier and a target identifier is generated for a transaction packet that comprises a header portion, a data portion and an end-to-end CRC portion. The transaction packet is then transmitted with the additional header from the ingress point to an egress point. At the egress point the additional header is removed. The additional header can further include a resource key. It can further include protection information such as a CRC.

Abstract: A PCI-Express interconnect device is provided. The device comprises: first and second upstream ports; first and second downstream ports; and a management entity. The management entity is operable to logically partition the device such that the first upstream port is operable to communicate with the first downstream port and the second upstream port is operable to communicate with the second downstream port. In one example, the management device is additionally operable to migrate the first downstream port from a partition in which the first downstream port is operable to communicate with the first upstream port into a partition in which the first downstream port is operable to communicate with the second upstream port.

Abstract: An infrastructure element can receive a first DMA request including a first address and the data, generate a meta request that comprises a resource key value and a doorbell address, and transmit the meta request via the infrastructure using the doorbell address. A remote DMA adapter can receive the meta request at the doorbell address and generate a remote direct memory access request message using the resource key, the first address and the data from the received meta request.

Abstract: One embodiment of the invention provides a method of operating a switch in a communications network. The method includes storing a set of routings for each port in the switch. When a packet is received at a port in the switch, the set of routings corresponding to the port at which the packet was received is accessed. A port from which to forward the packet from the switch is determined in accordance with the set of routings. The packet is then duly forwarded over the network from the port as determined by the routings.

Abstract: One embodiment of the invention provides a method of handling an incoming packet at a node in a network. The method includes associating at least one source identifier with a link into the node. A packet is then received at the node over the link, and a source identifier is extracted from the received packet. The extracted source identifier is compared with the source identifier(s) associated with the link in order to authenticate the packet.

Abstract: One embodiment of the invention provides a method of operating a switch with cut-through routing in a communications network. The method includes receiving a packet at a first port of the switch and determining a second port of the switch from which the received packet is to be forwarded out over the network. At least one authentication check is performed on the received packet. In general, the switch starts to forward the received packet from the second port prior to complete reception of the packet at the first port. However, this onward transmission is delayed until the authentication check has completed. This ensure that none of the received packet is forwarded from the second port if any authentication check fails.

Abstract: One embodiment of the invention provides apparatus and a method for handling an incoming packet at a port in a network. The port belongs to one or more partitions, and at least one service level is associated with each of the partitions. When a packet is received at the port, a partition key and a service level are extracted from the packet. The extracted partition key is used to determine the partition to which the received packet belongs. The service level extracted from the packet is then compared against the service level(s) associated with the determined partition in order to authenticate the packet. In another embodiment, the service level is correlated against a destination identifier from the packet instead of the partition key.