Method to optimize network data flows within a constrained system

A Peripheral Component Interconnect Express (PCIe) apparatus, configured to be installed in a multi-socket server platform including a local socket having a first host processor coupled to first host memory and a first PCIe interconnect and a remote socket having a second host processor coupled to second host memory and a second PCIe interconnect, the multi-socket server platform further including an InfiniBand (IB) Host Channel Adaptor (HCA) coupled to the first PCIe interconnect, the PCIe apparatus comprising:

• a processor;
• and
• memory, coupled to the processor, having software instructions stored thereon to perform operations when executed by the processor, enabling the apparatus to,
• allocate a first transmit queue in the memory;
• queue first data in the first transmit queue, the first data to be transferred outbound from the multi-socket server platform via the IB HCA;
• transfer the first data from the first transmit queue to a first proxy agent hosted by the first host processor, the first data to be forwarded by the first proxy agent to the IB HCA;
• and
• receive second data sent from the IB HCA,
• wherein the first and second data are transferred between the PCIe apparatus and the IB HCA along different paths, and wherein the PCIe apparatus is configured to be installed in each, one at a time, of a first PCIe slot coupled to the first PCIe interconnect and a second PCIe slot coupled to the second PCIe interconnect.

A tangible, non-transient machine-readable medium having software instructions stored thereon configured to be installed in a distributed manner on components in a multi-socket server platform including a local socket having a first host processor coupled to first host memory and a first Peripheral Component Interconnect Express (PCIe) interconnect and a remote socket having a second host processor coupled to second host memory and a second PCIe interconnect, the multi-socket server platform further including an InfiniBand (IB) Host Channel Adaptor (HCA) coupled to the first PCIe interconnect, the local socket comprising a local socket and the remote socket comprising a remote socket, the software instructions comprising:

• first software instructions configured to be run on each of the first and second host processor, including instructions for implementing a respective proxy agent on the first and second processors;
• and
• second software instructions configured to be run on a respective PCIe cards coupled to the first PCIe interconnect and the second PCIe interconnect,
• wherein the first software instructions and second software instructions are configured to facilitate data transfers including,
• a first data transfer of first data from a first PCIe card to the IB HCA, the first data being transferred from the first PCIe card to a first proxy agent, and then forwarded from the first proxy agent to the IB HCA;
• and
• a second data transfer of second data from the IB HCA to a second PCIe card, the second data being transferred from the first PCIe card to a second proxy agent, and then forwarded from the second proxy agent to the IB HCA.