Aggregating small remote memory access requests

A computer-implemented method, comprising:

• receiving, by a destination NIC from a source NIC, a plurality of packets, wherein a first packet comprises memory-operation requests previously aggregated and stored in a first queue of a first plurality of queues, each queue of the first plurality of queues corresponding to a respective destination NIC, and each respective packet of the received plurality of packets associated with a common destination NIC and a destination core;
• buffering asynchronously, by the destination NIC, the memory-operation requests of the first packet into a second plurality of queues based on a destination core associated with each request, the second plurality of queues comprising per-destination core queues, and wherein a number of the second plurality of queues is based on a number of destination cores;
• determining, by the destination NIC, that a total size of the memory-operation requests in a second queue of the second plurality of queues reaches a second predetermined threshold;
• and
• responsive to determining that the total size of the memory-operation requests in the second queue reaches the second predetermined threshold, aggregating the requests in the second queue into a second packet destined for the destination core.

A system comprising a source network interface card (NIC) and a destination network interface card (NIC), the source NIC comprising first integrated circuits to:

• receive, from a plurality of source cores, a stream of commands, a respective command comprising a first plurality of memory-operation requests;
• buffer asynchronously the memory-operation requests into a first queue of a first plurality of queues based on a destination NIC associated with each request, each queue of the first plurality of queues corresponding to a respective destination NIC, and wherein a number of the first plurality of queues is based on a number of destination NICs with which the source NIC is capable of communicating;
• determine whether a total size of the memory-operation requests in the first queue reaches a first predetermined threshold;
• and
• responsive to determining that the total size of the memory-operation requests in the first queue reaches the first predetermined threshold, aggregate the requests in the first queue into a first packet destined for the destination NIC.