Directed acyclic graph template for data pipeline

A method for use with a computing system, the method comprising:

• receiving a directed acyclic graph (DAG) template specifying a data pipeline of a plurality of processing stages, wherein the DAG template is received from a program via a DAG application program interface (API) included in a DAG runtime environment;
• for each processing stage of the plurality of processing stages, selecting a respective processing device of a plurality of communicatively linked processing devices;
• determining a routing sequence between the plurality of processing devices according to the DAG template;
• transmitting one or more input packets encoding the plurality of processing stages to the respective processing devices selected for the processing stages as specified by the routing sequence;
• in response to transmitting the one or more input packets, receiving, from a processing device of the plurality of processing devices, one or more output packets encoding a processing result of the data pipeline;
• and
• executing a processing device shell of a plurality of processing device shells at a host server device, wherein executing the processing device shell includes:
• receiving an intermediate packet from a first processing device of the plurality of processing devices;
• and
• conveying the intermediate packet to a second processing device of the plurality of processing devices, wherein the second processing device is specified by an intermediate packet header of the intermediate packet.

A computing system comprising:

• a plurality of processing devices, wherein each processing device is a central processing unit (CPU), a graphical processing unit (GPU), an application-specific integrated circuit (ASIC), or a field-programmable gate array (FPGA);
• a processor configured to:
• receive a directed acyclic graph (DAG) template specifying a data pipeline of a plurality of processing stages, wherein the DAG template is received from a program via a DAG application program interface (API) included in a DAG runtime environment;
• for each processing stage of the plurality of processing stages, select a respective processing device of the plurality of processing devices;
• determine a routing sequence between the plurality of processing devices according to the DAG template;
• and
• execute the plurality of processing stages at the respective processing devices selected for those processing stages in an order specified by the routing sequence, wherein:
• executing the plurality of processing stages includes, at a processing device shell of a plurality of processing device shells:
• receiving an intermediate packet from a first processing device of the plurality of processing devices;
• and
• conveying the intermediate packet to a second processing device of the plurality of processing devices;
• and
• the second processing device is specified by an intermediate packet header of the intermediate packet.