21.2 Simple Timeout and Retransmission Example

Let's first look at the retransmission strategy used by TCP. We'll establish a connection, send some data to verify that everything is OK, disconnect the cable, send some more data, and watch what TCP does:

 bsdi %  telnet svr4 discard  Trying 140.252.13.34...     Connected to svr4.     Escape character is '^]'.  hello, world   send this line normally   and hi   disconnect cable before sending this line  Connection closed by foreign host.  output when TCP gives up after 9 minutes

Figure 21.1 shows the tcpdump output. (We have removed all the type-of-service information that is set by bsdi. )

Figure 21.1. Simple example of TCP's timeout and retransmission.

Lines 1, 2 , and 3 correspond to the normal TCP connection establishment. Line 4 is the transmission of "hello, world" (12 characters plus the carriage return and linefeed ), and line 5 is its acknowledgment. We then disconnect the Ethernet cable from svr4.

Line 6 shows "and hi" being sent. Lines 7-18 are 12 retransmissions of that segment, and line 19 is when the sending TCP finally gives up and sends a reset.

Examine the time difference between successive retransmissions: with rounding they occur 1, 3, 6, 12, 24, 48, and then 64 seconds apart. We'll see later in this chapter that the first timeout is actually set for 1.5 seconds after the first transmission. (The reason it occurs 1.0136 seconds after the first transmission, and not exactly 1.5 seconds, was explained in Figure 18.7.) After this the timeout value is doubled for each retransmission, with an upper limit of 64 seconds.

This doubling is called an exponential backoff. Compare this to the TFTP example in Section 6.5, where every retransmission occurred 5 seconds after the previous.

The time difference between the first transmission of the packet (line 6 at time 24.480) and the reset (line 19 at time 566.488) is about 9 minutes. Modern TCP's are persistent when trying to send data!

On most implementations this total timeout value is not tunable. Solaris 2.2 allows the administrator to change this (the tcp_ip_abort_interval variable in Section E.4) and its default is only 2 minutes, not the more common 9 minutes.