Using cache coherent FPGAS to accelerate remote access

A device for assisting an application dependent on remote memory of a remote host to run on a local host, the device comprising:

• a first circuit configured to:
• receive a notice of a cache miss caused by the application running on the local host, wherein cache misses occur due to page tables of the running application being programmed to indicate that all pages of the running application are present in a local memory of the local host when at least one page of the running application is not present in the local memory so that memory accesses by the running application cause cache misses instead of page faults;
• obtain a cache line satisfying the cache miss from the remote host responsible for remote memory containing the cache line using a translation table and a network connection to the remote host;
• copy the obtained cache line into the local memory;
• and
• provide the obtained cache line to the running application;
• and
• a second circuit configured to:
• monitor cache coherence events caused by the running application;
• and
• provide the notice of the cache miss to the first circuit.

A computer system comprising:

• a CPU residing in a first CPU socket of the system;
• a first memory coupled to the CPU, the CPU running a hypervisor and an application residing in the first memory;
• a device residing in a second CPU socket of the system;
• a second memory coupled to the device;
• wherein the hypervisor is configured to indicate to the system that pages of the running application are all present in the second memory when at least one page of the running application is not present in the second memory so that memory accesses by the running application cause cache misses instead of page faults;
• and
• wherein the device is configured to:
• receive a notice of a cache miss caused by the running application;
• obtain a cache line satisfying the cache miss from a remote host responsible for a remote memory containing the cache line using a translation table and a network connection to the remote host;
• copy the obtained cache line into the second memory;
• and
• provide the obtained cache line to the running application.