Cumulative TCP congestion control

• US 20070070906A1
• Filed: 09/28/2005
• Published: 03/29/2007
• Est. Priority Date: 09/28/2005
• Status: Active Grant

• Current Assignee: NetApp, Inc.
• Original Assignee: Network Appliance Incorporated (NetApp, Inc.)
• Inventors: Thakur, Manish
• Granted Patent: US 7,719,967 B2
• Time in Patent Office: Days
• US Class Current: 370/235
• 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: H04L 69/163 In-band adaptation of TCP data exchange; In-band control procedures

Abstract

Methods and apparatus, including computer program products, providing cumulative TCP (Transmission Control Protocol) congestion control. One method includes establishing multiple TCP connections between a transmitting host and a receiving host for sending data from the transmitting host to the receiving host; and using one congestion window on the transmitting host to perform congestion control for the multiple TCP connections as a whole.

Claims

1. A method comprising:

establishing a plurality of Transmission Control Protocol (TCP) connections between a transmitting host and a receiving host for sending data from the transmitting host to the receiving host; and

using one congestion window on the transmitting host to perform congestion control for the plurality of TCP connections as a whole.

2. The method of claim 1, wherein the plurality of TCP connections includes all of the TCP connections for sending data from the transmitting host to the receiving host.

3. The method of claim 1, wherein the plurality of TCP connections does not include all of the TCP connections for sending data from the transmitting host to the receiving host.

4. The method of claim 1, wherein the congestion window has a size and using one congestion window on the transmitting host to perform congestion control comprises:

determining the number of TCP segments to send across the plurality of TCP connections from the transmitting host to the receiving host at a next transmit time based at least in part on the size of the congestion window; and

sending the determined number of TCP segments from the transmitting host to the receiving host at the next transmit time.

5. The method of claim 1, wherein each of the plurality of TCP connections has its own send buffer, the method further comprising:

performing flow control for each of the plurality of TCP connections to determine a respective maximum number of TCP segments to transfer from the respective send buffers of the plurality of TCP connections into a common send queue;

transferring at most the determined respective maximum number of TCP segments from the respective send buffers of the plurality of TCP connections into the common send queue; and

performing congestion control on TCP segments in the common send queue.

6. The method of claim 5, further comprising:

calculating a fair share of the congestion window for one of the plurality of TCP connections; and

limiting the respective maximum number of TCP segments to transfer from the send buffer of the one of the plurality of TCP connections into the common send queue to the calculated fair share.

7. The method of claim 6, wherein calculating a fair share of the congestion window for the one TCP connection includes multiplying the size of the congestion window by a ratio of the number of TCP segments that can be transferred from the send buffer of the one TCP connection based solely on the results of flow control for the one TCP connection to the number of TCP segments that can be transferred from the send buffers of all of the plurality of TCP connections based solely on the results of flow control for the plurality of TCP connections.

8. The method of claim 1, further comprising:

creating the congestion window with an initial size based on a maximum segment size (MSS) when the first TCP connection of the plurality of TCP connections is established.

9. The method of claim 1, further comprising:

decreasing the size of the congestion window if the size of the congestion window is greater than a maximum number of bytes of in-transit data across all of the plurality of TCP connections over two or more average round-trip times between the transmitting host and the receiving host.

10. A computer program product, tangibly embodied in an information carrier, the computer program product being operable to cause a data processing apparatus to perform operations comprising:

establishing a plurality of TCP connections between a transmitting host and a receiving host for sending data from the transmitting host to the receiving host; and

using one congestion window on the transmitting host to perform congestion control for the plurality of TCP connections as a whole.

11. The computer program product of claim 10, wherein the congestion window has a size and using one congestion window on the transmitting host to perform congestion control comprises:

determining the number of TCP segments to send across the plurality of TCP connections from the transmitting host to the receiving host at a next transmit time based at least in part on the size of the congestion window; and

sending the determined number of TCP segments from the transmitting host to the receiving host at the next transmit time.

12. The computer program product of claim 11, the operations further comprising:

performing flow control for each of the plurality of TCP connections to determine a respective maximum number of TCP segments to transfer from the respective send buffers of the plurality of TCP connections into a common send queue;

transferring at most the determined respective maximum number of TCP segments from the respective send buffers of the plurality of TCP connections into the common send queue; and

performing congestion control on TCP segments in the common send queue.

13. The computer program product of claim 12, the operations further comprising:

calculating a fair share of the congestion window for one of the plurality of TCP connections; and

limiting the respective maximum number of TCP segments to transfer from the send buffer of the one of the plurality of TCP connections into the common send queue to the calculated fair share.

14. The computer program product of claim 10, the operations further comprising:

decreasing the size of the congestion window if the size of the congestion window is greater than a maximum number of bytes of in-transit data across all of the plurality of TCP connections over two or more average round-trip times between the transmitting host and the receiving host.

15. The computer program product of claim 14, wherein decreasing the size of the congestion window includes decreasing the size of the congestion window to the maximum number of bytes of in-transit data across all of the plurality of TCP connections over the two or more average round-trip times between the transmitting host and the receiving host.

16. The program product of claim 10, the operations further comprising:

decreasing the size of the congestion window if a new TCP segment is sent from the transmitting host to the receiving host across any of the plurality of TCP connections and the new TCP segment is unacknowledged by the receiving host within a retransmission timeout interval.

17. The program product of claim 16, wherein decreasing the size of the congestion window includes decreasing the size of the congestion window by a fraction that is inversely proportional to the number of the plurality of TCP connections.

18. A method comprising:

establishing a plurality of TCP connections between a transmitting host and a receiving host for sending data from the transmitting host to the receiving host, wherein each of the plurality of TCP connections has its own send buffer;

performing flow control for each of the plurality of TCP connections to determine a respective maximum number of TCP segments to transfer from the respective send buffers of the plurality of TCP connections into a common send queue;

transferring at most the determined respective maximum number of TCP segments from the respective send buffers of the plurality of TCP connections into the common send queue; and

using one congestion window to perform congestion control on TCP segments in the send queue.

19. A transmitting host, comprising:

a network interface to at least one network, the network interface being operable to transmit and receive information to and from a receiving host communicatively coupled to the network; and

a network stack operable to control communication over the network interface, the network stack being operable to establish a plurality of TCP connections between a transmitting host and a receiving host for sending data from the transmitting host to the receiving host, wherein each of the plurality of TCP connections has its own send buffer, to perform flow control for each of the plurality of TCP connections to determine a respective maximum number of TCP segments to transfer from the respective send buffers of the plurality of TCP connections into a common send queue, to transfer at most the determined respective maximum number of TCP segments from the respective send buffers of the plurality of TCP connections into the common send queue, and to use one congestion window to perform congestion control on TCP segments in the send queue.

20. The transmitting host of claim 19, wherein the network stack is further operable to:

decrease the size of the congestion window if the size of the congestion window is greater than a maximum number of bytes of in-transit data across all of the plurality of TCP connections over two or more average round-trip times between the transmitting host and the receiving host.