Hardware/software co-optimization to improve performance and energy for inter-VM communication for NFVs and other producer-consumer workloads
Methods and apparatus implementing Hardware/Software co-optimization to improve performance and energy for inter-VM communication for NFVs and other producer-consumer workloads. The apparatus include multi-core processors with multi-level cache hierarchies including and L1 and L2 cache for each core and a shared last-level cache (LLC). One or more machine-level instructions are provided for proactively demoting cachelines from lower cache levels to higher cache levels, including demoting cachelines from L1/L2 caches to an LLC. Techniques are also provided for implementing hardware/software co-optimization in multi-socket NUMA architecture system, wherein cachelines may be selectively demoted and pushed to an LLC in a remote socket. In addition, techniques are disclosure for implementing early snooping in multi-socket systems to reduce latency when accessing cachelines on remote sockets.
1. A processor, configured to be implemented in a multi-socketed computer system employing a Non-Uniform Memory Access (NUMA) architecture in which multiple instances of the processor are to be installed, each socket including a processor and local system memory, comprising:
a plurality of cores, each having at least one associated cache occupying a respective level in a cache hierarchy;
a last level cache (LLC), communicatively coupled to the plurality of cores;
a memory controller, communicatively coupled to the plurality of cores, configured to support access to external system memory when the processor is installed in the computer system; and
a socket-to-socket interconnect interface configured to facilitate communication between processors installed in different sockets;
wherein each of the caches associated with a core, and the LLC include a plurality of cacheline slots for storing cacheline data, and wherein the processor is further is configured to support a machine instruction that when executed by a core on a processor operating as a local socket causes the processor to:
demote a cacheline to the LLC cache on the local socket; and
push a copy of the cacheline to an LLC of a processor installed in a remote socket, wherein the cacheline is pushed via the socket-to-socket interconnect interface.
2. The processor of claim 1 , wherein execution of the machine instruction causes a write-back of the cacheline to system memory.
3. The processor of claim 1 , wherein the processor includes a third level (L3) cache and a fourth level (L4) cache comprising the LLC.
4. The processor of claim 1 , wherein execution of the instruction causes the processor to release ownership of the cacheline in the processor operating as the local socket.
5. The processor of claim 1 , wherein execution of the instruction causes the processor to mark the cacheline as (S)hared.
6. The processor of claim 1 , wherein execution of the instruction causes the processor to:
write-back the cacheline from the LLC on the local socket to system memory on the local socket; and
release the cacheline in the LLC on the local socket.
7. The processor of claim 1 , wherein the processor is further configured to:
receive a cacheline that has been pushed from a processor operating in a remote socket via the socket-to-socket interconnect interface; and
store the cacheline in its LLC.
8. A method implemented on a processor including a plurality of cores, each having at least one associated cache occupying a respective layer in a cache hierarchy, and a last level cache (LLC) that is shared among the plurality of cores, the processor further including a memory controller to access system memory when the processor is installed in a multi-socketed computer system employing a Non-Uniform Memory Access (NUMA) architecture in which multiple instances of the processor are installed, each socket including a processor and local system memory, and the processor further includes a socket-to-socket interconnect interface configured to facilitate communication between processors installed in different sockets, the method comprising:
demoting a cacheline to the LLC cache on the local socket via execution of an associated cacheline demotion machine instruction on a processor core; and
pushing a copy of the cacheline to an LLC of a processor installed in a remote socket, wherein the cacheline is pushed via the socket-to-socket interconnect interface.
9. The method of claim 8 , wherein execution of the machine instruction causes a write-back of the cacheline to system memory.
10. The method of claim 8 , further comprising releasing ownership of the cacheline in the processor operating as the local socket.
11. The method of claim 8 , further comprising:
performing a write-back of the cacheline from the LLC on the local socket to system memory on the local socket; and
releasing the cacheline in the LLC on the local socket.
12. The method of claim 8 , further comprising:
receiving a cacheline that has been pushed from a processor operating in a remote socket via the socket-to-socket interconnect interface; and
storing the cacheline in the LLC of the processor to which the cacheline is pushed.