Ensuring predictable and quantifiable networking performance

• US 8,804,523 B2
• Filed: 06/21/2012
• Issued: 08/12/2014
• Est. Priority Date: 06/21/2012
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Kim, Changhoon, Greenberg, Albert G., Dabagh, Alireza, Khalidi, Yousef A., Bansal, Deepak, Kandula, Srikanth
• Primary Examiner(s): LY, ANH VU H
• Application Number:
  US13/530,043
• Publication Number:
  US 20130343191A1
• Time in Patent Office: 782 Days
• Field of Search: None
• US Class Current: 370/235
• CPC Class Codes: G06F 2009/45595 Network integration; Enabling network access in virtual machine instances
• CPC Class Codes: G06F 9/45558 Hypervisor-specific management and integration aspects
• CPC Class Codes: H04L 47/11 Identifying congestion
• CPC Class Codes: H04L 47/12 Avoiding congestion; Recovering from congestion
• CPC Class Codes: H04L 47/263 Rate modification at the source after receiving feedback
• CPC Class Codes: H04L 47/762 triggered by the network
• CPC Class Codes: H04L 47/822 Collecting or measuring resource availability data
• CPC Class Codes: H04L 49/70 Virtual switches
• CPC Class Codes: H04L 69/16 Implementation or adaptation of Internet protocol [IP], of transmission control protocol [TCP] or of user datagram protocol [UDP]
• CPC Class Codes: Y02D 30/50 in wire-line communication networks, e.g. low power modes or reduced link rate

Abstract

The ensuring of predictable and quantifiable networking performance. Embodiments of the invention combine a congestion free network core with a hypervisor based (i.e., edge-based) throttling design to help insure quantitative and invariable subscription bandwidth rates. A lightweight shim layer in a hypervisor can adaptively throttle the rate of VM-to-VM traffic flow. A receiving hypervisor can detect congestion and communicate back to sending hypervisors that rates are to be regulated. In response, sending hypervisors can reduce transmission rate to mitigate congestion at the receiving hypervisor. In some embodiments, the principles are extended to any message processors communicating over a congestion free network.

Claims

1. In a cloud computing system that includes a plurality of virtual machines providing subscription services for one or more users, with the plurality of virtual machines each including a message processor that communicates packets comprising data flows to one or more hypervisors each having an adaptive rate limiter, with each hypervisor connected to a network core that communicates the data flows to at least one computer system including one or more processors and system memory, and a message processor for processing the data flows received from the network core, a method for addressing network congestion arising from one or more of the virtual machines, the method comprising the following acts:

accessing at a congestion detector provided at the computer system a subscription bandwidth for the message processor, the subscription bandwidth indicating a quantitative and invariant minimum bandwidth for the message processor;

receiving at the congestion detector one or more data flows from the network core;

calculating at the congestion detector the combined bandwidth of the one or more received data flows;

detecting at the congestion detector the onset of congestion at the message processor of the computer system by determining that the combined bandwidth that is calculated is within a specified threshold;

identifying at the congestion detector at least one message processor of one of the virtual machines associated with the one or more data flows as a violator of the subscription bandwidth in response to detecting the onset of congestion;

determining at the congestion detector the extent of the violation;

formulating at the congestion detector at least one feedback message that serves to rate limit the at least one violating message processor; and

sending from the congestion detector the feedback message onto the network core for delivery to at least one of the adaptive rate limiters.

2. The method as recited in claim 1, wherein at least one data flow comprises a Transmission Control Protocol (TCP) and at least one other data flow comprises a protocol different than Transmission Control Protocol (TCP).

3. The method as recited in claim 1, wherein said computer system is a virtual machine associated with a hypervisor that includes said congestion detector.

4. The method as recited in claim 2, wherein the at least one feedback message is an instruction to reduce the bandwidth of the data flow of the at least one adaptive rater limiter to which it is sent.

5. The method as recited in claim 4, wherein the at least one feedback message instruction is an instruction to reduce the bandwidth in accordance with an additive increase/multiplicative decrease (“

AIMD”

) algorithm such that data flow traffic using the different protocol is reduced based on the additive increase/multiplicative decrease (“

AIMD”

) algorithm.

6. The method as recited in claim 2, wherein the other protocol comprises User Datagram Protocol (“

UDP”

).

7. The method as recited in claim 1, wherein the act of detecting the onset of congestion at the computing system comprises an act of a software meter detecting the onset of congestion at the computing system.

8. The method as recited in claim 1, wherein the act of detecting the onset of congestion at the computing system comprises an act of using Explicit Congestion Notification (“

ECN”

) to detect the onset of congestion at the computing system.

9. The method as recited in claim 3, wherein the act of detecting the onset of congestion is performed by a virtual switch within the hypervisor that includes the congestion detector.

10. The method as recited in claim 1, wherein the adaptive rate limiter are each included within a virtual switch of at hypervisor.

11. The method as recited in claim 4, wherein the at least one feedback message instruction is an instruction for reducing the bandwidth of the one or more data flows in accordance with an additive increase/multiplicative decrease (“

AIMD”

) algorithm.

12. One or more computer hardware storage devices having stored thereon computer executable instructions that, when executed at a processor, implement a computer-controlled method for addressing network congestion arising from one or more of virtual machines in a cloud computing system, the virtual machines providing subscription services for one or more users, and the virtual machines each including a message processor that communicates packets comprising data flows to one or more hypervisors each having an adaptive rate limiter, with each hypervisor connected to a network core that communicates the data flows to at least one computer system including one or more processors and system memory, and a message processor for processing the data flows received from the network core, and the computer-controlled method comprising the following acts:

accessing at a congestion detector provided at the computer system a subscription bandwidth for the message processor, the subscription bandwidth indicating a quantitative and invariant minimum bandwidth for the message processor;

receiving at the congestion detector one or more data flows from the network core;

calculating at the congestion detector the combined bandwidth of the one or more received data flows;

detecting at the congestion detector the onset of congestion at the message processor of the computer system by determining that the combined bandwidth that is calculated is within a specified threshold;

identifying at the congestion detector at least one message processor of one of the virtual machines associated with the one or more data flows as a violator of the subscription bandwidth in response to detecting the onset of congestion;

determining at the congestion detector the extent of the violation; and

formulating at the congestion detector at least one feedback message that serves to rate limit the at least one violating message processor; and

sending from the congestion detector the feedback message onto the network core for delivery to at least one of the adaptive rate limiters.

13. One or more hardware storage devices as defined in claim 12, further comprising computer-executable instructions that, when executed, cause the computer system to:

receive at least one data flow of Transmission Control Protocol (TCP) traffic; and

receive at least one other data flow of traffic using another protocol, different than Transmission Control Protocol (TCP).

14. One or more hardware storage devices as defined in claim 12, further comprising computer-executable instructions that, when executed, cause the computer system to formulate the feedback message for reducing data flow bandwidth in accordance with an additive increase/multiplicative (“

AIMD”

) decrease algorithm.

15. One or more hardware storage devices as defined in claim 12, further comprising computer-executable instructions that, when executed, cause the computer system to use software metering to detect the onset of congestion at the computing system.

16. One or more hardware storage devices as defined in claim in claim 12, further comprising computer-executable instructions that, when executed, cause the computer system to use Explicit Congestion Notification (“

ECN”

) to detect the onset of congestion at the computing system.

17. One or more hardware storage devices as defined in claim 12, wherein the message processor of the computer system is a virtual machine served by a hypervisor which included the congestion detector, and wherein the one or more hardware storage devices further comprise computer-executable instructions that cause the act of detecting the onset of congestion to be performed by a virtual switch within the hypervisor that includes the congestion detector.

18. A cloud computing system comprising:

one or more virtual machines providing subscription services for one or more users, the virtual machines each including a message processor that communicates packets comprising data flows;

at least one or more hypervisors associated with each virtual machine for controlling allocation of host processes and resources to each virtual machine, each hypervisor having an adapative rate limiter component associated with the hypervisor;

a network core which sends and receives the data flows to or from one or more hypervisors;

a computer system including one or more processors and system memory, and a message processor for processing the data flows received from the network core;

a congestion detector component associated with said computer system; and

one or more computer storage devices having stored thereon computer executable instructions that, when executed at the processor of said computing system, implement a computer-controlled method for addressing network congestion arising from one or more of the virtual machines, and wherein the computer-controlled method is comprised of the following acts;

accessing at the congestion detector component provided at the computer system a subscription bandwidth for the computer system'"'"'s message processor, the subscription bandwidth indicating a quantitative and invariant minimum bandwidth for the message processor of the computer system;

receiving at the congestion detector component one or more data flows from the network core;

calculating at the congestion detector component the combined bandwidth of the one or more received data flows;

detecting at the congestion detector component the onset of congestion at the message processor of the computer system by determining that the combined bandwidth that is calculated is within a specified threshold;

identifying at the congestion detector at least one message processor of one of the virtual machines associated with the one or more data flows as a violator of the subscription bandwidth in response to detecting the onset of congestion;

determining at the congestion detector component the extent of the violation; and

formulating at the congestion detector component at least one feedback message that serves to rate limit the at least one violating message processor; and

sending from the congestion detector the feedback message onto the network core for delivery to at least one of the adaptive rate limiters.

19. The system of claim 18, wherein the congestion detector is configured to detect the onset of congestion using one or more of software metering and Explicit Congestion Notification (“

ECN”

).

20. The system of claim 18, wherein each adaptive rate limiter is configured to reduce the rate of one or more data flows in accordance with a congestion avoidance algorithm in response to receiving the feedback message, and the algorithm reduces the rate of User Datagram Protocols (“

UDP”

) data flows in accordance with an additive increase/multiplicative (“

AIMD”

) decrease algorithm.

21. A system as defined in claim 18, wherein said computer system is a virtual machine associated with a hypervisor that includes said congestion detector component.