[e2e] TCP in outer space

[Fred Baker]

At 01:15 PM 4/16/2001 -0700, Jonathan Stone wrote:
>I buy the 4 RTT-> 3 RTT argument. Could you clarify what you said
>about how this example reflects the rest of your samples-- that it's
>both the median and the shortest?

this exchange involves 13 messages. The very shortest involved one message 
- a DNS request for which there was no response. That wasn't a great 
example of an HTTP request.

The shortest TCP exchange was six messages:

---------------------------------------------------------------------
13:21:36.300572 fred-hm-dhcp1.cisco.com.2537 > 204.29.171.10.80:
        S 585709:585709(0) win 8192 <mss 1460> (DF) [tos 0x10] (ttl 128, id 
60682)
13:21:36.337219 204.29.171.10.80 > fred-hm-dhcp1.cisco.com.2537:
         S 1380537855:138 0537855(0) ack 585710 win 8760 <mss 1460> (DF) 
(ttl 112, id 14349)
---------------------------------------------------------------------
13:21:36.341485 fred-hm-dhcp1.cisco.com.2537 > 204.29.171.10.80:
         . ack 1 win 8760 (DF) [tos 0x10] (ttl 128, id 61450)
13:21:36.343153 fred-hm-dhcp1.cisco.com.2537 > 204.29.171.10.80:
         P 1:551(550) ack 1 win 8760 (DF) [tos 0x10] (ttl 128, id 61706)
---------------------------------------------------------------------
13:21:36.519975 204.29.171.10.80 > fred-hm-dhcp1.cisco.com.2537:
         P 1:307(306) ack 551 win 8210 (DF) (ttl 112, id 19725)
13:21:36.523251 fred-hm-dhcp1.cisco.com.2537 > 204.29.171.10.80:
         R 586260:586260(0) win 0 (DF) [tos 0x10] (ttl 128, id 61962)
---------------------------------------------------------------------

I'll be happy if someone can tell me what in the world this was; it doesn't 
appear to have a graceful close, but it does appear to contain an 
acknowledged data exchange in each direction. From a timing perspective, it 
is 3 round trips. I had three instances of such a session, and four 
instances of sessions with seven messages.

This is the shortest that also attempted a graceful close, but it only 
closes one direction - the other direction is summarily reset and the data 
offered is not acknowledged:

---------------------------------------------------------------------
13:11:22.301629 fred-hm-dhcp1.cisco.com.1971 > rip.itc.Virginia.EDU.80:
         S 583291:583291(0) win 8192 <mss 1460> (DF) [tos 0x10] (ttl 128, 
id 44795)
13:11:22.401453 rip.itc.Virginia.EDU.80 > fred-hm-dhcp1.cisco.com.1971:
         S 2287361714:2287361714(0) ack 583292 win 16384 <mss 512> (ttl 40, 
id 3485)
---------------------------------------------------------------------
13:11:22.408590 fred-hm-dhcp1.cisco.com.1971 > rip.itc.Virginia.EDU.80:
         . ack 1 win 8192 (DF) [tos 0x10] (ttl 128, id 45307)
13:11:22.409960 fred-hm-dhcp1.cisco.com.1971 > rip.itc.Virginia.EDU.80:
         P 1:453(452) ack 1 win 8192 (DF) [tos 0x10] (ttl 128, id 45563)
13:11:22.629749 rip.itc.Virginia.EDU.80 > fred-hm-dhcp1.cisco.com.1971:
         P 1:326(325) ack 453 win 16384 (ttl 40, id 3527)
---------------------------------------------------------------------
13:11:22.630994 fred-hm-dhcp1.cisco.com.1971 > rip.itc.Virginia.EDU.80:
         R 583744:583744(0) win 0 (DF) [tos 0x10] (ttl 128, id 45819)
13:11:22.632375 rip.itc.Virginia.EDU.80 > fred-hm-dhcp1.cisco.com.1971:
         F 326:326(0) ack 453 win 16384 (ttl 40, id 3528)
13:11:22.632532 fred-hm-dhcp1.cisco.com.1971 > rip.itc.Virginia.EDU.80:
         R 583744:583744(0) win 0 (ttl 128, id 46331)
---------------------------------------------------------------------

Interestingly, I had 11 instances of this exchange, and 33 and 39 
respectively of similar nine and ten message exchanges. No idea what that 
is about.

Here is the shortest complete session, in the sense of message count, in 
that it carried a request and a response, and properly closing in both 
directions (eg, both ends sent data and both ends had reason to know that 
their peer received it). Note, however, that it occupies *five* round trips!

---------------------------------------------------------------------
13:11:07.586812 fred-hm-dhcp1.cisco.com.1965 > 206.253.217.8.80:
         S 583285:583285(0) win 8192 <mss 1460> (DF) [tos 0x10] (ttl 128, 
id 33787)
13:11:07.638509 206.253.217.8.80 > fred-hm-dhcp1.cisco.com.1965:
         S 630920270:630920270(0) ack 583286 win 17520 <mss 1460> (ttl 47, 
id 3286)
---------------------------------------------------------------------
13:11:07.648165 fred-hm-dhcp1.cisco.com.1965 > 206.253.217.8.80:
         . ack 1 win 8760 (DF) [tos 0x10] (ttl 128, id 34299)
13:11:07.649719 fred-hm-dhcp1.cisco.com.1965 > 206.253.217.8.80:
         P 1:500(499) ack 1 win 8760 (DF) [tos 0x10] (ttl 128, id 34555)
13:11:07.843564 206.253.217.8.80 > fred-hm-dhcp1.cisco.com.1965:
         P 1:1456(1455) ack 500 win 17520 (ttl 47, id 3405)
---------------------------------------------------------------------
13:11:07.845358 206.253.217.8.80 > fred-hm-dhcp1.cisco.com.1965:
         F 1456:1456(0) ack 500 win 17520 (ttl 47, id 3441)
13:11:07.845529 fred-hm-dhcp1.cisco.com.1965 > 206.253.217.8.80:
         . ack 1457 win 7305 (DF) [tos 0x10] (ttl 128, id 35835)
---------------------------------------------------------------------
13:11:08.104627 fred-hm-dhcp1.cisco.com.1965 > 206.253.217.8.80:
         P 500:999(499) ack 1457 win 7305 (DF) [tos 0x10] (ttl 128, id 36603)
13:11:08.106239 fred-hm-dhcp1.cisco.com.1965 > 206.253.217.8.80:
         F 999:999(0) ack 1457 win 7305 (DF) [tos 0x10] (ttl 128, id 36859)
---------------------------------------------------------------------
13:11:08.190612 image4.go2net.com.80 > fred-hm-dhcp1.cisco.com.1965:
         R 630921727:630921727(0) win 0 (ttl 46, id 3893)
13:11:08.193635 206.253.217.8.80 > fred-hm-dhcp1.cisco.com.1965:
         R 630921727:630921727(0) win 0 (ttl 47, id 3892)
---------------------------------------------------------------------

The one I sent earlier was the median in my small sample. It carried an 
entire transaction using one round trip to get the request out and one 
round trip (overlapped some with the other) to get the response. It used 
one round trip to set up the connection, and one round trip to gracefully 
close it. It was shortest of the complete sessions in that it occupied four 
(one could say 3.5, as two of those overlapped in time) round trips. It was 
median in message count.