Rate proportional scheduling to reduce packet loss in virtualized network function chains

1. A system, comprising:

a host computing device comprising a host processor and a host memory; and

a hypervisor stored in the host memory and executed by the host processor to manage the execution of a first virtual machine and a second virtual machine by the host computing device, wherein:

the first virtual machine comprises, a first virtualized network function, a receive queue, and a congestion monitor configured to at least:

detect congestion in the receive queue; and

send a pause signal to a rate controller executed by the second virtual machine; and

the second virtual machine comprises a second virtualized network function comprising a transmit queue, and a rate controller configured to at least:

receive the pause signal; and

cause the second virtualized network function to pause processing of packets to reduce congestion in the receive queue; wherein:

the first virtualized network function is implemented using a first virtualized accelerator card allocated to the first virtual machine, the first virtualized accelerator card virtualizing at least a portion of a host accelerator card of the host computing device;

the second virtualized network function is implemented using a second virtualized accelerator card allocated to the second virtual machine, the second virtualized accelerator card virtualizing at least a portion of the host accelerator card of the host computing device; and

pausing the processing of packets by the second virtualized network function causes the host computing device to reallocate a time-slice for the host accelerator card from the second virtualized network function to the first virtualized network function.

2. The system of claim 1 , wherein

the congestion monitor of the first virtual machine is further configured to at least:

detect starvation in the receive queue; and

send a resume signal to the rate controller executed by the second virtual machine; and

the rate controller of the second virtual machine is further configured to at least:

receive the resume signal; and

cause the second virtualized network function to resume processing of packets to reduce starvation in the receive queue.

3. The system of claim 2 , wherein the congestion monitor is configured to detect starvation in the receive queue of the first virtualized network function by at least:

identifying a number of packets in the receive queue; and

determining that the number of packets in the receive queue is less than a threshold number of packets.

4. The system of claim 1 , wherein the congestion monitor is configured to detect congestion in the receive queue by at least:

identifying, by the congestion monitor, a number of packets in the receive queue; and

determining, by the congestion monitor, that the number of packets in the receive queue of the first virtualized network function exceeds a threshold number of packets.

5. The system of claim 1 , wherein the host accelerator card is capable of general purpose computation.

6. The system of claim 1 , wherein the first virtualized accelerator card is a first virtualized graphics processing unit (GPU), the second virtualized accelerator card is a second virtualized GPU, and the host accelerator card is a host GPU.

7. The system of claim 1 , wherein the first virtualized network function is at least one of a virtual private network (VPN) endpoint, an intrusion detection system, or a routing system.

8. A method, comprising:

detecting congestion, by a congestion monitor executed by a first virtual machine executed by a host computing device, in a receive queue associated with a first virtualized network function implemented by a first virtual machine;

sending, by the congestion monitor executed by the first virtual machine executed by the host computing device, a pause signal to a rate controller executed by a second virtual machine executed by the host computing device;

receiving, by the rate controller executed by the second virtual machine executed by the host computing device, the pause signal; and

pausing, by the rate controller, processing of packets by a second virtualized network function implemented by the second virtual machine to reduce congestion in the receive queue of the first virtualized network function; wherein:

the first virtualized network function is implemented using a first virtualized accelerator card allocated to the first virtual machine, the first virtualized accelerator card virtualizing at least a portion of a host accelerator card of the host computing device;

the second virtualized network function is implemented using a second virtualized accelerator card allocated to the second virtual machine, the second virtualized accelerator card virtualizing at least a portion of the host accelerator card of the host computing device; and

pausing the processing of packets by the second virtualized network function causes the host computing device to reallocate a time-slice for the host accelerator card from the second virtualized network function to the first virtualized network function.

9. The method of claim 8 , further comprising:

detecting starvation, by the congestion monitor executed by the first virtual machine, in the receive queue;

sending, by the congestion monitor executed by the first virtual machine, a resume signal to the rate controller executed by the second virtual machine;

receiving, by the rate controller executed by the second virtual machine, the resume signal; and

resuming, by the rate controller, processing of packets with the second virtualized network function implemented by the second virtual machine to reduce starvation in the receive queue.

10. The method of claim 9 , wherein detecting starvation, by the congestion monitor executed by the first virtual machine, in the receive queue further comprises:

identifying, by the congestion monitor, a number of packets in the receive queue; and

determining, by the congestion monitor, that the number of packets in the receive queue is less than a threshold number of packets.

11. The method of claim 8 , wherein detecting congestion, by the congestion monitor executed by the first virtual machine, in the receive queue further comprises:

identifying, by the congestion monitor, a number of packets in the receive queue; and

determining, by the congestion monitor, that the number of packets in the receive queue exceeds a threshold number of packets.

12. The method of claim 8 , wherein the host accelerator card is capable of general purpose computation.

13. The method of claim 8 , wherein the first virtualized accelerator card is a first virtualized graphics processing unit (GPU), the second virtualized accelerator card is a second virtualized GPU, and the host accelerator card is a host GPU.

14. The method of claim 8 , wherein the first virtualized network function is at least one of a virtual private network (VPN) endpoint, an intrusion detection system, or a routing system.

15. A non-transitory, computer-readable medium, comprising machine-readable instructions that, when executed by a processor of a host computing device, cause the computing device to at least:

detect congestion, with a congestion monitor executed by a first virtual machine executed by the host computing device, in a receive queue associated with a first virtualized network function implemented by a first virtual machine;

send, with the congestion monitor executed by the first virtual machine executed by the host computing device, a pause signal to a rate controller executed by a second virtual machine executed by the host computing device;

receive, by the rate controller executed by the second virtual machine executed by the host computing device, the pause signal; and

pause, by the rate controller, processing of packets by a second virtualized network function implemented by the second virtual machine to reduce congestion in the receive queue of the first virtualized network function; wherein:

the first virtualized network function is implemented using a first virtualized accelerator card allocated to the first virtual machine, the first virtualized accelerator card virtualizing at least a portion of a host accelerator card of the host computing device;

the second virtualized network function is implemented using a second virtualized accelerator card allocated to the second virtual machine, the second virtualized accelerator card virtualizing at least a portion of the host accelerator card of the host computing device; and

pausing the processing of packets by the second virtualized network function causes the host computing device to reallocate a time-slice for the host accelerator card from the second virtualized network function to the first virtualized network function.

16. The non-transitory, computer-readable medium of claim 15 , wherein the machine-readable instructions, when executed by the processor, further cause the host computing device to at least:

detect starvation, by the congestion monitor executed by the first virtual machine, in the receive queue;

send, by the congestion monitor executed by the first virtual machine, a resume signal to the rate controller executed by the second virtual machine;

receive, by the rate controller executed by the second virtual machine, the resume signal; and

resume, by the rate controller, processing of packets with the second virtualized network function implemented by the second virtual machine to reduce starvation in the receive queue.

17. The non-transitory, computer-readable medium of claim 16 , wherein detecting starvation, by the congestion monitor executed by the first virtual machine, in the receive queue further comprises causing the host computing device to at least:

identify, by the congestion monitor, a number of packets in the receive queue; and

determine, by the congestion monitor, that the number of packets in the receive queue is less than a threshold number of packets.

18. The non-transitory, computer-readable medium of claim 15 , wherein detecting congestion, by the congestion monitor executed by the first virtual machine, in the receive queue further comprises:

identifying, by the congestion monitor, a number of packets in the receive queue; and

determining, by the congestion monitor, that the number of packets in the receive queue exceeds a threshold number of packets.

19. The non-transitory, computer-readable medium of claim 15 , wherein the host accelerator card is capable of general purpose computation.

20. The non-transitory, computer-readable medium of claim 15 , wherein the first virtualized accelerator card is a first virtualized graphics processing unit (GPU), the second virtualized accelerator card is a second virtualized GPU, and the host accelerator card is a host GPU.