Exabyte-scale data processing system

1. A data processing system, comprising:

a plurality of processing units, each processing unit having one or more processor cores;

a plurality of dynamic random access memory (DRAM) devices, each DRAM device being coupled to one of the processing unit via a corresponding memory bus interface;

a plurality of memory roots, each memory root being associated with one of the processing units, wherein each memory root comprises

one or more branches, wherein each memory root further comprises a memory controller to provide an interface between the processing unit and one or more branches associated with the memory root, and

a plurality of memory leaves to store data, wherein each of the branches is associated with one or more of the memory leaves and to provide access to the data stored therein, wherein each of the memory leaves comprises one or more solid state memory devices; and

a memory fabric coupled to each of the branches of each memory root to allow each branch to access data stored in any of the memory leaves associated with any one of the branches and to allow each of the branches of each memory root to communicate with any of the branches;

wherein each of the branches further comprises a set of queues, wherein the set of queues associated with each of the branch comprises:

a submission queue to store one or more commands issued by a corresponding processing unit by each of the branches; and

a completion queue to store a result of executing the commands dispatched from the submission queue;

wherein each of the branches is to provide access to data stored in local memory leaves to a remote processing unit of the plurality of processing units via a remote branch of the remote processing unit over the memory fabric, without having to invoke a local processing unit and in response to a command issued by the remote processing unit and stored in the submission queue associated with the local memory leaves, a local branch of the branches executes the command to access data stored in a remote memory leaf of the memory leaves by communicating the command with the remote branch by selecting a first path via a first branch of the one or more branches, or a second path via a second branch of the one or more branches if the first path fails, without having to invoke an associated processing unit of a branch of the selected path, and

wherein the memory bus interfaces of the processing units are coupled to each other to allow the local processing unit accessing data stored in a remote DRAM device of the remote processing unit, without having to invoke the remote processing unit.

2. The system of claim 1 , wherein in response to the command received by the remote branch, the remote branch executes the command to access the data stored in the remote memory leaf and returns a result of accessing the data back to the local branch.

3. The system of claim 2 , wherein the local branch stores the result of accessing the data in the completion queue accessed by a local memory controller.

4. The system of claim 1 , wherein each processing unit is coupled to its associated memory root via a peripheral component interface express (PCIe) link.

5. The system of claim 1 , wherein the memory fabric comprises an Ethernet connection or an Ethernet network.

6. The system of claim 1 , wherein each of the processor cores represents a thread processor, and wherein each thread processor is to execute a specific thread.

7. The system of claim 1 , wherein memory bus interfaces of the processing units are coupled to each other via a shared memory bus.

8. The system of claim 1 , wherein memory bus interfaces of the processing units are coupled to each other via an interconnect.

9. The system of claim 1 , wherein memory bus interfaces of the processing units are coupled to each other via a network.