Adaptive load balancing in packet processing

1. A method for forwarding packets in a network device, the method comprising:

receiving a packet;

mapping the packet of a flow to a bucket, wherein the bucket is associated with a packet processing thread from a plurality of packet processing threads;

determining whether the packet processing thread is oversubscribed;

responsive to determining that the packet processing thread is not oversubscribed,

performing the following:

setting a load balancing flag of the packet of the flow to a first value indicating that the packet is directly mapped to the packet processing thread instead of being load balanced over one of the plurality of packet processing threads, and

mapping the packet of the flow to the packet processing thread;

responsive to determining that the packet processing thread is oversubscribed, performing the following:

setting the load balancing flag of the packet of the flow to a second value indicating that the packet is load balanced over one of the plurality of packet processing threads instead of being directly mapped to the packet processing thread,

distributing the packet to one of the plurality of packet processing threads based on a predefined load balancing scheme, and

processing the packet of the flow in the one of the plurality of packet processing threads; and

forwarding the packet according to a predetermined order, wherein the predetermined order is based on a position of the packet relative to other packets of the flow at their receipt in the network device, wherein the packet of the flow is received after a preceding packet of the flow at the network device, and forwarding the packet of the flow further includes:

responsive to determining that the packet of the flow has a load balancing flag set to the first value, outputting the packet of the flow, and

responsive to determining that the packet of the flow has a load balancing flag set to the second value, performing the following:

determining whether the preceding packet of the flow is still to be processed by one of the plurality of packet processing threads,

responsive to determining that the preceding packet of the flow is still to be processed, storing the packet in a buffer, and

outputting the packet of the flow following the processing and output of the preceding packet of the flow.

2. The method of claim 1 , wherein responsive to determining that the packet processing thread is oversubscribed, further evaluating a state of the bucket, wherein the state of the bucket is indicative of a packet rate.

3. The method of claim 2 , wherein distributing the packet is further based on the evaluated state of the bucket.

4. The method of claim 2 , wherein the state of the bucket includes one or a combination of:

a counter value for the bucket, wherein the counter value is indicative of the number of packets in the bucket; and

an exponential weighted moving average (EWMA) value of the packet rate for the bucket.

5. The method of claim 1 , further comprising:

assigning a sequence number to the packet prior to distributing the packet to one of the plurality of packet processing threads, and

wherein forwarding the packet is performed according to the assigned sequence number.

6. The method of claim 1 , wherein determining whether the packet processing thread is oversubscribed includes:

evaluating a length of a queue of the packet processing thread; and

determining whether the length of the queue is above a predefined threshold.

7. The method of claim 6 , wherein the length of the queue is an exponential weighted moving average (EWMA) queue length.

8. The method of claim 1 , wherein mapping the packet to the bucket is performed according to a hashing function.

9. A non-transitory machine-readable storage medium that provides instructions, which when executed by a processor of a network device, cause said processor to perform operations comprising:

receiving a packet;

mapping the packet of a flow to a bucket, wherein the bucket is associated with a packet processing thread from a plurality of packet processing threads;

determining whether the packet processing thread is oversubscribed;

responsive to determining that the packet processing thread is not oversubscribed,

performing the following:

setting a load balancing flag of the packet of the flow to a first value indicating that the packet is directly mapped to the packet processing thread instead of being load balanced over one of the plurality of packet processing threads, and

mapping the packet of the flow to the packet processing thread;

responsive to determining that the packet processing thread is oversubscribed,

performing the following:

setting the load balancing flag of the packet of the flow to a second value indicating that the packet is load balanced over one of the plurality of packet processing threads instead of being directly mapped to the packet processing thread,

distributing the packet to one of the plurality of packet processing threads based on a predefined load balancing scheme, and

processing the packet of the flow in the one of the plurality of packet processing threads; and

forwarding the packet according to a predetermined order, wherein the predetermined order is based on a position of the packet relative to other packets of the flow at their receipt in the network device, wherein the packet of the flow is received after a preceding packet of the flow at the network device, and forwarding the packet of the flow further includes:

responsive to determining that the packet of the flow has a load balancing flag set to the first value, outputting the packet of the flow, and

responsive to determining that the packet of the flow has a load balancing flag set to the second value, performing the following:

determining whether the preceding packet of the flow is still to be processed by one of the plurality of packet processing threads,

responsive to determining that the preceding packet of the flow is still to be processed, storing the packet in a buffer, and

outputting the packet of the flow following the processing and output of the preceding packet of the flow.

10. The non-transitory machine-readable storage medium of claim 9 , wherein responsive to determining that the packet processing thread is oversubscribed, further evaluating a state of the bucket, wherein the state of the bucket is indicative of a packet rate.

11. The non-transitory machine-readable storage medium of claim 10 , wherein distributing the packet is further based on the evaluated state of the bucket.

12. The non-transitory machine-readable storage medium of claim 10 , wherein the state of the bucket includes one or a combination of:

a counter value for the bucket, wherein the counter value is indicative of the number of packets in the bucket; and

an exponential weighted moving average (EWMA) value of the packet rate for the bucket.

13. The non-transitory machine-readable storage medium of claim 9 , wherein the operations further comprise:

assigning a sequence number to the packet prior to distributing the packet to one of the plurality of packet processing threads, and

wherein forwarding the packet is performed according to the assigned sequence number.

14. The non-transitory machine-readable storage medium of claim 9 , wherein determining whether the packet processing thread is oversubscribed includes:

evaluating a length of a queue of the packet processing thread; and

determining whether the length of the queue is above a predefined threshold.

15. The non-transitory machine-readable storage medium of claim 14 , wherein the length of the queue is an exponential weighted moving average (EWMA) queue length.

16. The non-transitory machine-readable storage medium of claim 9 , wherein mapping the packet to the bucket is performed according to a hashing function.