Abstract: An apparatus in an example comprises an RDMA (Remote Direct Memory Access) network adapter located at an interface of a requester node coherency domain that sends an RDMA read request to a responder node coherency domain to flush into the responder node coherency domain data previously sent by the RDMA network adapter.

Abstract: A local RDMA (Remote Direct Memory Access) network adapter that comprises a simplex switchless connection with a counterpart QP on a remote RDMA network adapter in an example is selected for a Queue Pair (QP). An apparatus in an example comprises a requester RDMA (Remote Direct Memory Access) session fail-over coordinator on a coherency domain that adds a session fail-over header to a front of a data payload sent by an RDMA computer program from the coherency domain. Upon termination, of an RDMA adapter not coherent with the coherency domain and in a communication path of the data payload, before delivery to the RDMA computer program of a receive completion notification for the data payload, the session fail-over header is employable by a responder RDMA session fail-over coordinator to reconstruct the receive completion notification.

Abstract: A first one of multiple embedded processing elements (12-14) in a computer (10) receives a delivery packet (124) that is formatted in accordance with a delivery protocol and includes (i) an encapsulated payload packet (136) that is formatted in accordance with a payload protocol and (ii) a delivery packet header (134) including routing information. In response to a determination that it is not the destination for the delivery packet (124), the first processing element (14) sends the delivery packet (124) from the first processing element (14) to a second one of the processing elements based on the routing information. In response to a determination that it is the destination for the delivery packet (124), the first processing element (14) decapsulates the payload packet (136) from the delivery packet (124) and processes the decapsulated payload packet (136).

Abstract: An endpoint device (14) is registered in association with a host memory address in response to receipt of a request for a notification of a change in content state of the host memory address from the endpoint device (14). In response to a change in content state of the host memory address, a notification that the host memory address has changed content state is sent to the endpoint device (14). In response to receipt of the notification by the endpoint device (14), semantics associated with a change of content state of the host memory address by a data schema (42) is determined and an action is performed by the endpoint device (14) in accordance with the determined semantics.

Abstract: Systems and methods of fast synchronization failure detection in distributed databases are disclosed. An example of a method includes receiving a digest of a database stored at a sending node in a network, the digest broadcast by the sending node to N number of nodes in the network. The method also includes generating a digest of a database stored at a receiving node in the network. The method also includes comparing the generated digest to the received digest. The method also includes issuing a lost synchronization signal by the receiving node when the comparison indicates a change in the database stored at the sending node.

Abstract: A cooling system for a blade enclosure is disclosed. The cooling system comprises a thermal bus bar (TBB) 1220 positioned in the middle of the blade enclosure. The TBB 122 has a front face and a back face. When blades are inserted into the blade enclosure, a heat transfer plate 584 on the blade makes thermal contact with either the front or back face of the TBB 122. The TBB 122 is cooled, thereby cooling the blades.

Abstract: A cooling system for a computer blade is disclosed. The cooling system comprises a main printed circuit (PC) board with at least one component mounted on a top side of the main PC board. A heat transfer plate is located at the first end of the main PC board. An airflow divider is mounted on, and is perpendicular with, the top side of the main PC board. The airflow divider runs in an axis generally parallel with the first side of the main PC board. A lid is coupled to the main PC board thereby enclosing the main PC board, the heat transfer plate and the airflow divider. The lid encloses a first airflow channel running along the first side of the main PC board and a second airflow channel running along the second side of the main PC board. A fan is located on top side of the main PC board in the first airflow channel. The fan is configured to re-circulate air from the second airflow channel through the first airflow channel, past the heat transfer plate, and then back into the second airflow channel.

Abstract: A method includes producing units of work with at least one source application instance (AI) at a source device and consuming units of work with at least one destination AI at a destination device. A first reliable datagram service is established, with a first source and destination resource (SDR), between the source device and the destination device. A second reliable datagram service is established, with a second SDR independent of the first SDR, between the source device and the destination device. A first unit of work stream is transmitted over a communication services/fabric with the first reliable datagram service, which guarantees strong ordering of the first unit of work stream received at the destination device. A second unit of work stream is transmitted over the communication services/fabric with the second reliable datagram service, which guarantees strong ordering of the second unit of work stream received at the destination device.

Abstract: An apparatus employs a work request list to access a memory device. The apparatus comprises an upper layer protocol that generates the work request list comprising a plurality of work requests, the work request list having an attribute that indicates the number of the plurality of work requests in the work request list. The apparatus additionally comprises an interface that is adapted to receive the work request list and individually enqueue the plurality of work requests.

Abstract: A computing device having partitions, and a method of communicating between partitions, are disclosed wherein each partition comprises at least one address area readable but not writable from the other of the at least two partitions. In one embodiment one partition sends to the other partition a request for information, which information is in the other partition in an address area not accessible to the one partition, the other partition copies the information to an address area accessible to the one partition, and the one partition reads the information from the accessible address area. In another embodiment the at least one accessible address area of each partition includes a data area and a consumer pointer indicating the position to which that partition has read the data area in another partition.

Abstract: A method includes producing units of work with at least one source application instance (AI) at a source device and consuming units of work with at least one destination AI at a destination device. A first reliable datagram service is established, with a first source and destination resource (SDR), between the source device and the destination device. A second reliable datagram service is established, with a second SDR independent of the first SDR, between the source device and the destination device. A first unit of work stream is transmitted over a communication services/fabric with the first reliable datagram service, which guarantees strong ordering of the first unit of work stream received at the destination device. A second unit of work stream is transmitted over the communication services/fabric with the second reliable datagram service, which guarantees strong ordering of the second unit of work stream received at the destination device.

Abstract: The disclosed embodiments relate to a method and apparatus for providing notification. The apparatus may comprise a plurality of completion queue handlers associated with a communication device. Each of the plurality of completion queue handlers may be associated with a plurality of processes. At least one completion queue may be associated with at least one of the plurality of completion queue handlers.

Abstract: A local RDMA (Remote Direct Memory Access) network adapter that comprises a simplex switchless connection with a counterpart QP on a remote RDMA network adapter in an example is selected for a Queue Pair (QP). An apparatus in an example comprises a requester RDMA (Remote Direct Memory Access) session fail-over coordinator on a coherency domain that adds a session fail-over header to a front of a data payload sent by an RDMA computer program from the coherency domain. Upon termination, of an RDMA adapter not coherent with the coherency domain and in a communication path of the data payload, before delivery to the RDMA computer program of a receive completion notification for the data payload, the session fail-over header is employable by a responder RDMA session fail-over coordinator to reconstruct the receive completion notification.

Abstract: An apparatus in an example comprises an RDMA (Remote Direct Memory Access) network adapter located at an interface of a requester node coherency domain that sends an RDMA read request to a responder node coherency domain to flush into the responder node coherency domain data previously sent by the RDMA network adapter.

Abstract: A reliable multicast service is operated between a source device and multiple destination devices participating in a multicast group. The source device includes a first source application instance (AI) producing a first unit of work stream, and communication services (CS). Each destination device in the multicast group includes CS, and at least one destination AI which consumes units of work. Communication services/fabric provide communication between the source device and the multiple destination devices. Multiple source and destination resources (SDRs) implement corresponding multiple reliable transport services between the source device and corresponding multiple destination devices in the multicast group for delivery of the first unit of work stream to the corresponding destination devices and guaranteeing strong ordering of the first unit of work stream received at the corresponding destination devices.

Abstract: A reliable datagram service is implemented with a source and destination resource (SDR). Source SDR resources, at a source device, multiplex units of work produced by at least one source application instance (AI) into a serial unit of work stream having units of work in a defined order and transmit the serial unit of work stream over a communication services/fabric. Destination SDR resources, at a destination device, receive the serial unit of work stream, demultiplex the serial unit of work stream into units of work provided to at least one destination AI, and provide a negative acknowledgement (NAK) for a unit of work received ahead of its defined order.

Abstract: A distributed computer system includes a source endnode including a source process which produces message data and a send work queue having work queue elements that describe the message data for sending. A destination endnode includes a destination process and a receive work queue having work queue elements that describe where to place incoming message data. A communication fabric provides communication between the source endnode and the destination endnode. An end-to-end context is provided at the source endnode and the destination endnode storing state information to ensure the reception and sequencing of message data sent from the source endnode to the destination endnode permitting reliable datagram service between the source endnode and the destination endnode.

Abstract: The disclosed embodiments may relate to memory window access and may include a memory window and plurality of queue pairs associated with a process. Each of the plurality of queue pairs may be associated with a memory window context that may have queue pair information. The memory window may be associated with a memory window context that includes a protection information field. Accordingly, access to memory window may be allowed if the queue pair information matches the protection information field.

Abstract: A distributed computer system includes links and routing devices coupled between the links and routing frames between the links. Each of the routing devices includes a congestion control mechanism for detecting congestion at the routing device and responding to detected congestion by gradually reducing an injection rate of frames routed from the routing device.

Abstract: The present invention relates to highly available sequence number generation with minimal latency. The present invention involves establishing a primary and secondary sequence number generators. A sequence number request is forwarded to the primary generator. The primary generator determines the next sequence number and forwards the response to the secondary generator. The secondary generator stores the value and forwards the value to the originating entity. Accordingly, the configuration avoids single point failure and is robust against malfunction.