1 / 24

XCP eXplicit Control Protocol

XCP eXplicit Control Protocol. A Linux implementation and analysis. What is XCP?. XCP was developed by Dina Katabi from MIT, and released as a draft RFC on October 17. 2004. XCP was conceived as an answer to the question: “Given our current knowledge of congestion control,

Télécharger la présentation

XCP eXplicit Control Protocol

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. XCPeXplicit Control Protocol A Linux implementation and analysis

  2. What is XCP? • XCP was developed by Dina Katabi from MIT, and released as a draft RFC on October 17. 2004. • XCP was conceived as an answer to the question: “Given our current knowledge of congestion control, how would we ideally like congestion control to work?” • XCP is a protocol developed in order to improve TCP’s congestion control algorithm. Especially in networks with a high bandwidth-delay product. • XCP takes a new approach congestion control by letting the routers in the network return explicit feedback back to the hosts.

  3. The Master Thesis • Create a working Linux implementation using the draft RFC directly. • Since the XCP protocol was to be implemented as a separate protocol layer, the implementation should not change any existing TCP/IP code. • Using this implementation to test the protocol in a real network. • Compare the results from the real Linux implementation with the original simulations by D. Katabi and other implementations.

  4. Why develop XCP? • TCP is 25 years old, and is designed to work in networks with much lower bandwidth-delay products than current networks. • TCP use packet drops to control transmission speed, and uses an AIMD (Arithmetic Increase, Multiplicative Decrease) algorithm to adjust the sending speed. • The higher the Bandwidth-Delay product, the less efficient TCP’s congestion control algorithms are. • Networks with high delays makes TCP react slower to packet drops. Additionally, bandwidth is reacquired more slowly in such networks.

  5. XCP’s solution • XCP allows the routers in the network to continuously adjust the sending speed of any participating hosts. • These adjustments are done by changing the contents of the packets (XCP header) transferred between the sender and receiver. • The feedback from routers are used by the sender to adjust the transfer speed to fit the routers current load. Router Router

  6. XCP protocol layer Application TCP XCP IP Link The TCP/XCP/IP stack • XCP introduces a 20 byte header between IP and TCP that carries information about the sender’s desired bandwidth and information about what the routers allow. • XCP requires all routers and hosts in the network to use the XCP protocol in order to work as intended. • The XCP protocol bases all adjustments of bandwidth on a per packet calculation. XCP routers does not need to separate between different data-flows, but make adjustments on each individual packet.

  7. The XCP header Version : 1 Format : 1 (Full header), 2 (Minimal header) Protocol : Next level protocol (6 for TCP) Length : 20 Unused : 0 RTT : Round Trip Time as measured from the sender in milliseconds Throughput: The current throughput in bytes/ms in use by the sender Delta_Throughput: The wanted change in throughput wanted by the sender, measured in bytes/second. Reverse_Feedback: The contents of the Delta_Throughput field when the message is returned back to the sender from the receiver. 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |Version| Format| Protocol | Length | unused | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | RTT | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Throughput | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Delta_Throughput | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Reverse_Feedback | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

  8. The XCP algorithm • The sender fills in the fields in the XCP header. The Delta_Throughput field contains the amount of bandwidth the sender wants to use if no congestion exist in the network. • The XCP routers checks incoming requests for bandwidth and downgrades the Delta_Throughput field if not enough bandwidth is available. • The receiver returns the Delta_Throughput back to the sender in the ACK message in the Reverse_Feedback field. • The reverse feedback is then used by the sender to adjust its sending rate accordingly. So that the sender will adjust to the available bandwidth in the network.

  9. An XCP network (simplified) 1 1 1 2 6 6 20 20 0 0 100 100 0 0 0 10.000.000 1 1 6 20 0 0 200.000 RTT Current Throughput Delta Throughput Feedback 100 0 200.000 0 Network RTT: 100 ms Router’s capacity: 200.000 B/s (available 200.000 B/s) Sender’s capacity: 10.000.000 B/s (available 10.000.000 B/s) Sender’s current throughput: 0 B/s (or 0 B/ms) Sender Router Receiver

  10. An XCP network (simplified) 1 1 1 2 6 6 20 20 0 0 100 100 200 0 0 9.800.000 1 1 6 20 0 0 0 RTT Current Throughput Delta Throughput Feedback 100 200 0 0 Network RTT: 100 ms Router’s capacity: 200.000 B/s (0 B/s is available) Sender’s capacity: 10.000.000 B/s (9.800.000 B/s is available) Sender’s current throughput: 200.000 B/s (or 200 B/ms) Sender Router Receiver

  11. XCP algorithms at sender When sending the current throughput is calculated using the formula: Delta throughput is calculated on a per packet basis using the following formula: The feedback is calculated using this formula:

  12. XCP algorithms at router • The router periodically (each average RTT milliseconds) calculates the load on itself, and adjusts how much bandwidth to allow for the following interval. • During the previous interval, the router collects data from the XCP and IP header of all passing packets (such as their RTT, throughput and size ). • This data is then used to calculate how adjustments to each packet’s delta throughput is done. • The XCP router basically allows 40% of available bandwidth to be distributed during the next interval.

  13. XCP router algorithms The XCP router uses the following formula to adjust how much bandwidth to allow for the next interval: The resulting feedback is then distributed using an AIMD (Arithmetic Increase, Multiplicative Decrease) algorithm. In the case where more bandwidth is available the XCP router will increase the bandwidth of all flows with the same amount. If too little bandwidth is available, the reduction for each XCP flow will be proportional with the flow’s current bandwidth usage.

  14. XCP router example Arithmetic Increase: A router having one flow using 1.000.000 B/s and ten flows using 10.000 B/s each is to distribute 440.000 B/s. This will give each flow additional 40.000 B/s, so the large flow will get 1.040.000 B/s and the small ones 50.000 B/s Multiplicative Decrease: A router having one flow using 1.000.000 B/s, and 10 flows using 10.000 B/s each, needs to reduce the throughput by 110.000 B/s (10%). The large flow will be reduced to 900.000 B/s while the 10 small flows to 9.000 B/s • The combination of these two algorithms allows each flow to acquire its fair share of the total available bandwidth in the network.

  15. XCP Linux implementation issues • XCP is based on floating point calculations, while the Linux kernel only allows integer arithmetic. • XCP needs to change TCP’s congestion window, thereby invalidating the TCP/IP stack protocol hierarchy. • In addition XCP needs to know about different TCP flows to work. Further invalidating the protocol hierarchy. • Hardware checksums in network cards failed, because of the addition of a new header between TCP and IP.

  16. XCP Performance Overestimation of congestion window, due to invalid feedback. • The throughput = cwnd/RTT formula leads to invalid XCP feedback during startup • XCP fails to improve startup speed compared to TCP, because of TCP’s reliance of the peer’s “advertised window”.

  17. XCP Router Queues vs. TCP • XCP manages to prevent router queues and packet drops on full duplex network.

  18. XCP bandwidth distribution • XCP manages to quickly distribute the available bandwidth fairly and smoothly between different flows. The bandwidth usage for each flow stays constant, and does not oscillate.

  19. XCP fails in half duplex mode • On half duplex network, the XCP router fails give correct feedback to the hosts. This causes queue buildup, and more oscillatory bandwidth usage.

  20. XCP fails if sender is bursty Inflated congestion window, as XCP fails to give negative feedback. • If the sender transfers data periodically, the XCP protocol fails to converge. This leads to large queue buildups, and packet drops. • Unfortunately for XCP, periodic transfers of data is very common in the real world.

  21. XCP fails if sender is bursty 2 • As the XCP protocol fails to reduce the congestion window, queues build up in the router. • This leads to packet drops, which XCP supposedly should prevent.

  22. Summary • XCP manages to deliver on its promises in situations where the XCP router can create a valid model of the network load. • When the XCP router fails to correctly predict its load, the protocol fails. These situations include half-duplex links, lack of buffer space, periodically sending applications, and overloaded hosts. • Implementing XCP as a separate protocol, does not give enough control over TCP. Using the congestion window is not exact enough in order to for XCP to work. • A feedback mechanism that works in all real life scenarios must be more advanced than the one currently in use by XCP.

  23. The Future • XCP is “revolutionary” in the sense it uses feedback from the network to improve congestion control. • XCP’s design prevents gradual introduction in Internet. • XCP has problems working in real life situations. • Demanding more work to be done by routers is probably not a good idea for faster networks. • The XCP routers base their distribution of bandwidth to all flows, on input from flows that is not verified.

  24. Questions?

More Related