1TCP protocol 2============ 3 4Last updated: 9 February 2008 5 6Contents 7======== 8 9- Congestion control 10- How the new TCP output machine [nyi] works 11 12Congestion control 13================== 14 15The following variables are used in the tcp_sock for congestion control: 16snd_cwnd The size of the congestion window 17snd_ssthresh Slow start threshold. We are in slow start if 18 snd_cwnd is less than this. 19snd_cwnd_cnt A counter used to slow down the rate of increase 20 once we exceed slow start threshold. 21snd_cwnd_clamp This is the maximum size that snd_cwnd can grow to. 22snd_cwnd_stamp Timestamp for when congestion window last validated. 23snd_cwnd_used Used as a highwater mark for how much of the 24 congestion window is in use. It is used to adjust 25 snd_cwnd down when the link is limited by the 26 application rather than the network. 27 28As of 2.6.13, Linux supports pluggable congestion control algorithms. 29A congestion control mechanism can be registered through functions in 30tcp_cong.c. The functions used by the congestion control mechanism are 31registered via passing a tcp_congestion_ops struct to 32tcp_register_congestion_control. As a minimum name, ssthresh, 33cong_avoid, min_cwnd must be valid. 34 35Private data for a congestion control mechanism is stored in tp->ca_priv. 36tcp_ca(tp) returns a pointer to this space. This is preallocated space - it 37is important to check the size of your private data will fit this space, or 38alternatively space could be allocated elsewhere and a pointer to it could 39be stored here. 40 41There are three kinds of congestion control algorithms currently: The 42simplest ones are derived from TCP reno (highspeed, scalable) and just 43provide an alternative the congestion window calculation. More complex 44ones like BIC try to look at other events to provide better 45heuristics. There are also round trip time based algorithms like 46Vegas and Westwood+. 47 48Good TCP congestion control is a complex problem because the algorithm 49needs to maintain fairness and performance. Please review current 50research and RFC's before developing new modules. 51 52The method that is used to determine which congestion control mechanism is 53determined by the setting of the sysctl net.ipv4.tcp_congestion_control. 54The default congestion control will be the last one registered (LIFO); 55so if you built everything as modules, the default will be reno. If you 56build with the defaults from Kconfig, then CUBIC will be builtin (not a 57module) and it will end up the default. 58 59If you really want a particular default value then you will need 60to set it with the sysctl. If you use a sysctl, the module will be autoloaded 61if needed and you will get the expected protocol. If you ask for an 62unknown congestion method, then the sysctl attempt will fail. 63 64If you remove a tcp congestion control module, then you will get the next 65available one. Since reno cannot be built as a module, and cannot be 66deleted, it will always be available. 67 68How the new TCP output machine [nyi] works. 69=========================================== 70 71Data is kept on a single queue. The skb->users flag tells us if the frame is 72one that has been queued already. To add a frame we throw it on the end. Ack 73walks down the list from the start. 74 75We keep a set of control flags 76 77 78 sk->tcp_pend_event 79 80 TCP_PEND_ACK Ack needed 81 TCP_ACK_NOW Needed now 82 TCP_WINDOW Window update check 83 TCP_WINZERO Zero probing 84 85 86 sk->transmit_queue The transmission frame begin 87 sk->transmit_new First new frame pointer 88 sk->transmit_end Where to add frames 89 90 sk->tcp_last_tx_ack Last ack seen 91 sk->tcp_dup_ack Dup ack count for fast retransmit 92 93 94Frames are queued for output by tcp_write. We do our best to send the frames 95off immediately if possible, but otherwise queue and compute the body 96checksum in the copy. 97 98When a write is done we try to clear any pending events and piggy back them. 99If the window is full we queue full sized frames. On the first timeout in 100zero window we split this. 101 102On a timer we walk the retransmit list to send any retransmits, update the 103backoff timers etc. A change of route table stamp causes a change of header 104and recompute. We add any new tcp level headers and refinish the checksum 105before sending. 106 107