Scaled-out transport as connection proxy for device-to-device communications

A network system comprising:

• a plurality of data processing units (DPUs) interconnected via a network fabric, wherein each DPU of the plurality of DPUs implements proxy logic for a system bus connection, and wherein the plurality of DPUs and the network fabric together operate as a single system bus connection proxy;
• a host device locally attached to a host unit interface of a first DPU of the plurality of DPUs via a first system bus connection;
• and
• a plurality of endpoint devices locally attached to host unit interfaces of one or more second DPUs of the plurality of DPUs via second system bus connections, wherein the proxy logic for the system bus connection configures each of the host unit interface of the first DPU and the host unit interfaces of the second DPUs to operate in one of more of an endpoint mode, a rootport mode, a switch upstream function mode, or a switch downstream function mode,
• wherein the first DPU is configured to, upon receipt of packets from the host device on the host unit interface of the first DPU and destined for a given endpoint device of the plurality of endpoint devices, establish a logical tunnel across the network fabric between the first DPU and one of the second DPUs to which the given endpoint device is locally attached, encapsulate the packets using a transport protocol, and send the encapsulated packets over the logical tunnel to the one of the second DPU, and
• wherein the one of the second DPUs is configured to, upon receipt of the encapsulated packets, extract the packets, and send the packets on a host unit interface of the one of the second DPUs to the given endpoint device.

A first data processing unit (DPU) integrated circuit comprising:

• a networking unit interconnected with a plurality of DPUs via a network fabric;
• a host unit comprising a host unit interface locally attached to a host device via a system bus connection;
• and
• at least one processing core configured to:
• execute proxy logic for a system bus connection, wherein the plurality of DPUs, including the first DPU integrated circuit, and the network fabric together operate as a single system bus connection proxy, and wherein the host unit interface is configured to provide access to the single system bus connection proxy operating as at least one of a virtual switch attached to one or more of a plurality of endpoint devices or a virtual device implemented as an abstraction of one or more of the plurality of endpoint devices, and
• upon receipt of packets from the host device on the host unit interface and destined for a given endpoint device of the plurality of endpoint devices, establish a logical tunnel across the network fabric between the first DPU integrated circuit and a second DPU integrated circuit of the plurality of DPUs to which the given endpoint device is locally attached, encapsulate the packets using a transport protocol including packing multiple system bus connection transaction layer packets (TLPs) into an Ethernet frame, and send the encapsulated packets over the logical tunnel to the second DPU integrated circuit.

A first data processing unit (DPU) integrated circuit comprising:

• a networking unit interconnected with a plurality of DPUs via a network fabric;
• a host unit comprising a host unit interface locally attached to a host device via a system bus connection;
• and
• at least one processing core configured to:
• execute proxy logic for a system bus connection, wherein the plurality of DPUs, including the first DPU integrated circuit, and the network fabric together operate as a single system bus connection proxy, wherein the host unit interface is configured to provide access to the single system bus connection proxy operating as at least one of a virtual switch attached to one or more of a plurality of endpoint devices or a virtual device implemented as an abstraction of one or more of the plurality of endpoint devices, and wherein the proxy logic for the system bus connection configures the host unit interface of the first DPU integrated circuit to operate in one or more of an endpoint mode, a rootport mode, a switch upstream function mode, or a switch downstream function mode, and
• upon receipt of packets from the host device on the host unit interface and destined for a given endpoint device of the plurality of endpoint devices, establish a logical tunnel across the network fabric between the first DPU integrated circuit and a second DPU integrated circuit of the plurality of DPUs to which the given endpoint device is locally attached, encapsulate the packets using a transport protocol, and send the encapsulated packets over the logical tunnel to the second DPU integrated circuit.

A method comprising:

• configuring, by a first data processing unit (DPU) of a plurality of DPUs interconnected via a network fabric and implementing proxy logic for a system bus connection, a host unit interface of the first DPU to operate in a first mode for a system bus connection by which the host unit interface is locally attached to a host device, wherein the plurality of DPUs and the network fabric together operate as a single system bus connection proxy, and wherein the host unit interface of the first DPU is configured to provide access to the single system bus connection proxy operating as at least one of a virtual switch attached to one or more of a plurality of endpoint devices or as a virtual device implemented as an abstraction of one or more of the plurality of endpoint devices;
• receiving, on the host unit interface of the first DPU, packets from the host device on the host unit interface, wherein the packets are destined for a given endpoint device of the plurality of endpoint devices;
• establishing a logical tunnel across the network fabric between the first DPU and a second DPU of the plurality of DPUs to which the given endpoint device is locally attached;
• encapsulating the packets using a transport protocol including packing multiple system bus connection transaction layer packets (TLPs) into an Ethernet frame;
• and
• sending the encapsulated packets over the logical tunnel to the second DPU.

A method comprising:

• configuring, by a first data processing unit (DPU) of a plurality of DPUs interconnected via a network fabric and implementing proxy logic for a system bus connection, a host unit interface of the first DPU to operate in a first mode for a system bus connection by which the host unit interface is locally attached to a host device, wherein the plurality of DPUs and the network fabric together operate as a single system bus connection proxy, wherein the host unit interface of the first DPU is configured to provide access to the single system bus connection proxy operating as at least one of a virtual switch attached to one or more of a plurality of endpoint devices or as a virtual device implemented as an abstraction of one or more of the plurality of endpoint devices, and wherein configuring the host unit interface of the first DPU to operate in the first mode for the system bus connection comprises configuring, by the proxy logic for the system bus connection, the host unit interface of the first DPU to operate in one or more of an endpoint mode, a rootport mode, a switch upstream function mode, or a switch downstream function mode;
• receiving, on the host unit interface of the first DPU, packets from the host device on the host unit interface, wherein the packets are destined for a given endpoint device of the plurality of endpoint devices;
• establishing a logical tunnel across the network fabric between the first DPU and a second DPU of the plurality of DPUs to which the given endpoint device is locally attached;
• encapsulating the packets using a transport protocol;
• and
• sending the encapsulated packets over the logical tunnel to the second DPU.