Computer architecture with disaggregated memory and high-bandwidth communication interconnects
Conventional high performance computer connections are electron-based systems, which require the memory packages to be as close as mechanically possible to the computation engine. Low power and high bandwidth communication, e.g. photonic, links can drastically change the architecture of high-performance computers by eliminating the bottlenecks in communication. A computer system comprises: a plurality of memory aggregation devices configured to retrieve data from and store data in a plurality of random access memory modules forming a unified contiguous memory address space disaggregated from a central processing unit; a plurality of computational devices configured for simultaneously launching a plurality of data signals including memory read and/or write requests for the data to the plurality of memory aggregation devices; and a plurality of communication links coupling each of the plurality of memory aggregation devices to each of the plurality of computational devices for transferring the data therebetween.
1. A computer system comprising:
a plurality of memory aggregation devices configured to retrieve data and store data from a plurality of random access memory modules;
a plurality of memory copy engines that launch a plurality of memory read and/or write requests for the data to the plurality of memory aggregation devices; and
a plurality of communication links, the plurality of communication links optically coupling each of the plurality of memory aggregation devices to each of the plurality of memory copy engines;
wherein the plurality of memory copy engines are configured to access greater than ⅔ of an available read/write bandwidth of the plurality of random access memory modules in parallel and uniformly.
2. The computer system according to claim 1 , wherein at least some of the plurality of memory copy engines are co-located with processing cores that are configured to compute data.
3. The computer system according to claim 1 , wherein at least some of the plurality of memory copy engines are co-located with bridge devices that are configured to communicate with other processors using a standard interface.
4. The computer system according to claim 1 , wherein greater than ⅔ of available random access memory in the plurality of random access modules is accessible via the plurality of memory aggregation devices accessible via the plurality of communication links.
5. The computer system according to claim 1 , wherein the communication links connected to each of the plurality of memory aggregation devices provide >⅔ of an available read/write bandwidth of the plurality of random access memory modules.
6. The computer system according to claim 1 , further comprising a switching system configured to provide any to any switching of data between the plurality of memory copy engines and the plurality of memory aggregation devices.
7. The computer system according to claim 6 , wherein a combination of the plurality of memory copy engines, the plurality of memory aggregation devices, and the switching system, provide a uniform memory to all of the plurality of memory copy engines.
8. The computer system according to claim 1 , wherein the plurality of memory copy engines are configured for memory interleaving, which is globally applied across all of the plurality of memory modules via the plurality of memory aggregation devices.
9. The computer system according to claim 8 , wherein a total random access memory capacity across all of the plurality of memory modules via the plurality of memory aggregation devices is greater than 10 TB.
10. The computer system according to claim 1 , wherein a total random access memory capacity is greater than 1 PB.
11. The computer system according to claim 1 , wherein the plurality of memory copy engines are configured to read/write multiple requests in parallel to all of the plurality of memory aggregation devices across a unified contiguous memory address space in the plurality of memory modules.
12. The computer system according to claim 1 , wherein a total random access memory capacity across all of the plurality of memory modules via the plurality of memory aggregation devices is greater than 10 TB.
13. A computer system comprising:
a plurality of memory aggregation devices configured to retrieve data and store data from a plurality of random access memory modules;
a plurality of memory copy engines that launch a plurality of memory read and/or write requests for the data to the plurality of memory aggregation devices; and
a plurality of communication links, the plurality of communication links optically coupling each of the plurality of memory aggregation devices to each of the plurality of memory copy engines; and
a switching system configured to provide any to any switching of data between the plurality of memory copy engines and the plurality of memory aggregation devices;
wherein a combination of the plurality of memory copy engines, the plurality of memory aggregation devices, and the switching system, provide a uniform memory to all of the plurality of memory copy engines.
14. The computer system according to claim 13 , wherein the plurality of memory copy engines are configured to access greater than 2% of an available read/write bandwidth of the plurality of random access memory modules in parallel and uniformly.
15. The computer system according to claim 13 , wherein a total random access memory capacity across all of the plurality of memory modules via the plurality of memory aggregation devices is greater than 10 TB.
16. The computer system according to claim 15 , wherein the total random access memory capacity is greater than 1 PB.
17. A computer system comprising:
a plurality of memory aggregation devices configured to retrieve data and store data from a plurality of random access memory modules;
a plurality of memory copy engines that launch a plurality of memory read and/or write requests for the data to the plurality of memory aggregation devices; and
a plurality of communication links, the plurality of communication links optically coupling each of the plurality of memory aggregation devices to each of the plurality of memory copy engines;
wherein the plurality of memory copy engines are configured for memory interleaving, which is globally applied across all of the plurality of memory modules via the plurality of memory aggregation devices; and
wherein a total random access memory capacity across all of the plurality of memory modules via the plurality of memory aggregation devices is greater than 10 TB.
18. A computer system comprising:
a plurality of memory aggregation devices configured to retrieve data and store data from a plurality of random access memory modules;
a plurality of memory copy engines that launch a plurality of memory read and/or write requests for the data to the plurality of memory aggregation devices; and
a plurality of communication links, the plurality of communication links optically coupling each of the plurality of memory aggregation devices to each of the plurality of memory copy engines;
wherein the plurality of memory copy engines are configured to access all of the plurality of memory modules via all of the plurality of memory aggregation devices in a unified, contiguous address space.
19. The computer system according to claim 18 , wherein the plurality of memory copy engines are configured to read/write multiple requests in parallel to all of the plurality of memory aggregation devices across a unified contiguous memory address space in the plurality of memory modules.
20. The computer system according to claim 18 , wherein a total random access memory capacity across all of the plurality of memory modules via the plurality of memory aggregation devices is greater than 10 TB.