Abstract: A system, method, and apparatus may provide one or more tangible, nontransitory computer-readable storage media having stored thereon executable instructions to instruct a processor to: stripe an outgoing network message into two or more pieces; send a first piece to a receiver via a first network interface card (NIC), and a second piece to the receiver via a second NIC; and upon determining that the receiver failed to receive a piece of the outgoing network message, replay the piece that the receiver failed to receive via a third NIC.

Abstract: An embodiment of a semiconductor package apparatus may include technology to embed one or more trigger operations in one or more messages related to collective operations for a neural network, and issue the one or more messages related to the collective operations to a hardware-based message scheduler in a desired order of execution. Other embodiments are disclosed and claimed.

Abstract: Examples include a method of managing storage for triggered operations. The method includes receiving a request to allocate a triggered operation; if there is a free triggered operation, allocating the free triggered operation; if there is no free triggered operation, recovering one or more fired triggered operations, freeing one or more of the recovered triggered operations, and allocating one of the freed triggered operations; configuring the allocated triggered operation; and storing the configured triggered operation in a cache on an input/output (I/O) device for subsequent asynchronous execution of the configured triggered operation.

Abstract: A system, method, and apparatus may provide one or more tangible, nontransitory computer-readable storage media having stored thereon executable instructions to instruct a processor to: stripe an outgoing network message into two or more pieces; send a first piece to a receiver via a first network interface card (NIC), and a second piece to the receiver via a second NIC; and upon determining that the receiver failed to receive a piece of the outgoing network message, replay the piece that the receiver failed to receive via a third NIC.

Abstract: Particular embodiments described herein provide for a system for enabling the communication of a large message using multiple network interface controllers (NICs). The system can be configured to determine that a message to communicate to a receiver over a network is above a threshold, determine a plurality of NICs to be used to communicate the message, create a manifest that includes an identifier of each of the plurality of NICs, and communicate the manifest to the receiver using a multi-unit message. In an example, the multi-unit message is communicated using a PUT command and the receiver can analyze the manifest and use a GET command to pull the message from the plurality of NICs.

Abstract: Methods, software, and systems for improved data transfer operations using overlapped rendezvous memory registration. Techniques are disclosed for transferring data between a first process operating as a sender and a second process operating as a receiver. The sender sends a PUT request message to the receiver including payload data stored in a send buffer and first and second match indicia. The first match indicia is used to determine whether the PUT request is expected or unexpected. If the PUT request is unexpected, an RMA GET operation is performed using the second matching indicia to pull data from the send buffer and write the data to a memory region in the user space of the process associated with the receiver. If the PUT request message is expected, the data payload with the PUT request is written to a receive buffer on the receiver determined using the first match indicia.

Abstract: Examples include a method of managing storage for triggered operations. The method includes receiving a request to allocate a triggered operation; if there is a free triggered operation, allocating the free triggered operation; if there is no free triggered operation, recovering one or more fired triggered operations, freeing one or more of the recovered triggered operations, and allocating one of the freed triggered operations; configuring the allocated triggered operation; and storing the configured triggered operation in a cache on an input/output (I/O) device for subsequent asynchronous execution of the configured triggered operation.

Abstract: Methods, software, and systems for improved data transfer operations using overlapped rendezvous memory registration. Techniques are disclosed for transferring data between a first process operating as a sender and a second process operating as a receiver. The sender sends a PUT request message to the receiver including payload data stored in a send buffer and first and second match indicia. Subsequent to or in conjunction with sending the PUT request message, the send buffer is exposed on the sender. The first match indicia is used to determine whether the PUT request is expected or unexpected. If the PUT request is unexpected, an RMA GET operation is performed using the second matching indicia to pull data from the send buffer and write the data to a memory region in the user space of the process associated with the receiver. The RMA GET operation may be retried one or more times in the event that the send buffer has yet to be exposed.

Abstract: Particular embodiments described herein provide for a system for enabling the communication of a message using a network interface controller (NICs) on a subnet. In an example, the system is applicable to hardware offload NICs such as those implementing the Portals protocol. The system can be configured to determine a NIC in a first subnet to be used to communicate a message, where the NIC is configured to comply with a message passing interface protocol, create a manifest that includes an identifier of the NICs and a subnet ID that identifies the first subnet, and communicate the manifest to the receiver.

Abstract: An embodiment of a semiconductor package apparatus may include technology to embed one or more trigger operations in one or more messages related to collective operations for a neural network, and issue the one or more messages related to the collective operations to a hardware-based message scheduler in a desired order of execution. Other embodiments are disclosed and claimed.

Abstract: Particular embodiments described herein provide for a system for enabling the communication of a large message using multiple network interface controllers (NICs). The system can be configured to determine that a message to communicate to a receiver over a network is above a threshold, determine a plurality of NICs to be used to communicate the message, create a manifest that includes an identifier of each of the plurality of NICs, and communicate the manifest to the receiver using a multi-unit message. In an example, the multi-unit message is communicated using a PUT command and the receiver can analyze the manifest and use a GET command to pull the message from the plurality of NICs.

Abstract: Technologies for targeted flow control recovery include a target computing device which detects whether resources of the target computing device are sufficient to process received messages from at least one source computing device and, in response to a determination that the resources are insufficient, (i) drops received message(s) from the affected source computing device(s) and (ii) determines whether to issue a targeted flow control recovery (i.e., at one of the source computing devices) or a global targeted flow control recovery (i.e., at all of the source computing devices) to instruct the source computing device(s) to stop transmitting messages to the target computing device.