Dynamically processing data message flows using different NUMA nodes of a processing system

1. A method for dynamically processing data message flows using different non-uniform memory access (NUMA) nodes of a processing system, each NUMA node comprising a local memory and a set of processors that can access data from local memories of other NUMA nodes, the method comprising:

at a load balancing application associated with a first NUMA node:

receiving data message flows destined for an endpoint application, wherein the data message flows are assigned to the first NUMA node to be forwarded to the endpoint application;

monitoring a central processing (CPU) usage of the first NUMA node to determine whether the CPU usage of the first NUMA node exceeds a particular threshold;

when the CPU usage of the first NUMA node exceeds the particular threshold, reassigning at least a subset of the data message flows to the second NUMA node for processing;

wherein unencapsulated flows are reassigned to the second NUMA node and encapsulated flows maintain assignment to the first NUMA node.

2. The method of claim 1 further comprising:

after reassigning the at least subset of data message flows to the second NUMA node:

forwarding the at least subset of data message flows to the second NUMA node to be processed;

receiving processed data message flows from the second NUMA node; and

forwarding the processed data message flows to the endpoint application.

3. The method of claim 1 , wherein the at least subset of data message flows comprises large data messages such that data message flows comprising small data messages maintain assignment to the first NUMA node.

4. The method of claim 1 , wherein the data message flows are initially assigned to the first NUMA node based on quality of service (QOS) parameters of the endpoint application.

5. The method of claim 4 , wherein the data message flows are initially assigned by a network administrator.

6. The method of claim 4 , wherein the data message flows are initially assigned by the load balancing application.

7. The method of claim 4 , wherein the data message flows are initially assigned to the first NUMA node because the data message flows are critical data message flows.

8. The method of claim 7 , wherein the critical data message flows are flows requiring a low latency.

9. The method of claim 7 , wherein the data message flows are a first set of data message flows, the endpoint application is a first endpoint application, and a second set of data message flows associated with a second endpoint application is assigned to the second NUMA node.

10. The method of claim 9 , wherein:

the first endpoint application is implemented by a first set of endpoint application instances,

the second endpoint application is implemented by a second set of endpoint application instances, and

the first and second sets of endpoint application instances execute on a set of one or more servers.

11. The method of claim 9 , wherein:

the first endpoint application is implemented by a first set of endpoint application instances executing on a first set of one or more servers, and

the second endpoint application is implemented by a second set of endpoint application instances executing on a second set of one or more servers.

12. The method of claim 9 , wherein the second set of data message flows is assigned to the second NUMA node because the second set of data message flows is a set of non-critical data message flows while the first set of data message flows is a set of critical data message flows.

13. The method of claim 12 , wherein the critical flows are flows requiring a low latency and the set of non-critical flows are flows not requiring a low latency.

14. The method of claim 1 , wherein the load balancing application is a first instance of a distributed load balancing application implemented by a plurality of instances operating on the different NUMA nodes.

15. The method of claim 1 further comprising, after the reassigning, storing a record, associating the at least subset of data message flows with the second NUMA node, in the local memory of the first NUMA node.

16. The method of claim 15 , wherein the record specifies, for each data message flow in the at least subset of data message flows, a flow identifier (ID) identifying the data message flow and a NUMA node ID identifying the second NUMA node.

17. The method of claim 1 , wherein the data message flows specify a fully qualified domain name (FQDN), specifying the endpoint application, as a destination of the data message flows.

18. The method of claim 17 , wherein the endpoint application is implemented by a set of endpoint application instances, and the FQDN specifies a particular endpoint application instance as the destination of the data message flows.

19. A non-transitory machine readable medium storing a program for execution by at least one processing unit for dynamically processing data message flows using different non-uniform memory access (NUMA) nodes of a processing system, each NUMA node comprising a local memory and a set of processors that can access data other local memories of other NUMA nodes, the program comprising sets of instructions for:

at a load balancing application associated with a first NUMA node:

receiving data message flows destined for an endpoint application, wherein the data message flows are assigned to the first NUMA node to be forwarded to the endpoint application;

monitoring a central processing (CPU) usage of the first NUMA node to determine whether the CPU usage of the first NUMA node exceeds a particular threshold;

when the CPU usage of the first NUMA node exceeds the particular threshold, reassigning at least a subset of the data message flows to the second NUMA node for processing;

wherein unencapsulated flows are reassigned to the second NUMA node and encapsulated flows maintain assignment to the first NUMA node.