Disaggregated physical memory resources in a data center
Examples may include sleds for a rack in a data center including physical compute resources and memory for the physical compute resources. The memory can be disaggregated, or organized into near and far memory. A first sled can comprise the physical compute resources and a first set of physical memory resources while a second sled can comprise a second set of physical memory resources. The first set of physical memory resources can be coupled to the physical compute resources via a local interface while the second set of physical memory resources can be coupled to the physical compute resources via a fabric.
1. A system for a data center comprising:
a rack comprising a plurality of sled spaces; and
a first sled coupled to the rack via a first one of the plurality of sled spaces, the first sled comprising:
at least one physical compute resource;
a first set of physical memory resources;
a local interface;
a first memory controller to couple the at least one physical compute resource to the first set of physical memory resources via the local interface;
a first fabric interface; and
a second memory controller to couple the at least one physical compute resource to a second set of physical memory resources via the first fabric interface;
a second sled coupled to the rack via a second one of the plurality of sled spaces, the second sled comprising:
the second set of physical memory resources, the second set of physical memory resources comprising byte-addressable write-in place non-volatile memory; and
a second fabric interface to couple the second set of physical memory resources to the at least one physical compute resource via a fabric, the first fabric interface, and the second memory controller.
2. The system of claim 1 , the first set of physical memory resources comprising volatile memory.
3. The system of claim 1 , the at least one physical compute resource comprising a central processing unit, a field programmable gate array, a graphics processing unit, or an application specific integrated circuit.
4. The system of claim 1 , the local interface a peripheral component interconnect express compliant standard.
5. The system of claim 1 , the fabric an optical fabric.
6. The system of claim 1 , wherein the local interface has lower latency than the fabric, greater bandwidth than the fabric, or both lower latency and greater bandwidth than the fabric.
7. The system of claim 1 , the at least one physical compute resource comprising a first physical compute resource and a second physical compute resource, the first memory controller to couple the first physical compute resource to the first set of physical memory resources via the local interface and to couple the first physical compute resource to the second set of physical memory resources via the first fabric interface, the second memory controller to couple the second physical compute resource to the second set of physical memory resources via the first fabric interface and to couple the second physical compute resource to the first set of physical memory resources via the local interface.
8. An apparatus for a compute resource sled, comprising:
at least one physical compute resource;
a first set of physical memory resources;
a local interface;
a first memory controller to couple the at least one physical compute resource to the first set of physical memory resources via the local interface;
a first fabric interface; and
a second memory controller to couple the at least one physical compute resource to a second set of physical memory resources via a fabric and the first fabric interface, the second set of physical memory resources comprising byte-addressable write-in place non-volatile memory.
9. The apparatus of claim 8 , the first set of physical memory resources comprising volatile memory.
10. The apparatus of claim 8 , the at least one physical compute resource comprising a central processing unit, a field programmable gate array, a graphics processing unit, or an application specific integrated circuit.
11. The apparatus of claim 8 , the local interface a peripheral component interconnect express compliant standard.
12. The apparatus of claim 8 , the fabric an optical fabric.
13. The apparatus of claim 8 , wherein the local interface has lower latency than the fabric, greater bandwidth than the fabric, or both lower latency and greater bandwidth than the fabric.
14. The apparatus of claim 8 , the at least one physical compute resource comprising a first physical compute resource and a second physical compute resource, the first memory controller to couple the first physical compute resource to the first set of physical memory resources via the local interface and to couple the first physical compute resource to the second set of physical memory resources via the first fabric interface, the second memory controller to couple the second physical compute resource to the second set of physical memory resources via the first fabric interface and to couple the second physical compute resource to the first set of physical memory resources via the local interface.
15. A method comprising:
coupling, via a first memory controller and a local interface, at least one physical compute resource to a first set of physical memory resources, the at least one physical compute resource and the first set of physical memory resources housed in a first sled of a data center; and
coupling, via a second memory controller and a fabric, the at least one physical compute resource to a second set of physical memory resources, the second set of physical memory resources housed in a second sled of the data center, the second set of physical memory resources comprising byte-addressable write-in place non-volatile memory.
16. The method of claim 15 , the first set of physical memory resources comprising volatile memory.
17. The method of claim 15 , the at least one physical compute resource comprising a central processing unit, a field programmable gate array, a graphics processing unit, or an application specific integrated circuit.
18. The method of claim 15 , the local interface a peripheral component interconnect express compliant standard.
19. The method of claim 15 , the fabric an optical fabric.
20. The method of claim 15 , wherein the local interface has lower latency than the fabric, greater bandwidth than the fabric, or both lower latency and greater bandwidth than the fabric.
21. The method of claim 15 , the at least one physical compute resource comprising a first physical compute resource and a second physical compute resource, the method comprising:
coupling, via the first memory controller and the local interface, the first physical compute resource to the first set of physical memory resources;
coupling, via the first memory controller and the fabric, the first physical compute resource to the second set of physical memory resources;
coupling, via the second memory controller and the local interface, the second physical compute resource to the first set of physical memory resources; and
coupling, via the second memory controller and the fabric, the second physical compute resource to the second set of physical memory resources.