Don't assume latency is symmetric

I wanted to document this before I forget. It always puzzled me that whenever I would do a traceroute to www.acis.ufl.edu that it would look like this

pierre@pdebian:~$ traceroute www.acis.ufl.edu

traceroute to www.acis.ufl.edu (128.227.56.252), 30 hops max, 60 byte packets

1 192.168.0.1 (192.168.0.1) 1.527 ms 1.936 ms 2.396 ms

2 10.4.128.1 (10.4.128.1) 10.125 ms 13.854 ms 19.332 ms

3 ip72-214-194-49.ga.at.cox.net (72.214.194.49) 20.313 ms 21.044 ms 21.776 ms

4 ip72-214-194-1.ga.at.cox.net (72.214.194.1) 20.010 ms 20.739 ms 21.460 ms

5 ip72-214-194-146.ga.at.cox.net (72.214.194.146) 19.444 ms 20.424 ms 24.656 ms

6 wsip-70-169-186-22.ga.at.cox.net (70.169.186.22) 24.888 ms 144.062 ms 33.085 ms

7 ssrb6c-nexus-msfc-1-v30-1.ns.ufl.edu (128.227.236.81) 14.307 ms 19.783 ms 20.015 ms

8 aerv1-core-msfc-1-v24-1.ns.ufl.edu (128.227.236.34) 20.233 ms 20.713 ms 20.948 ms

the main thing to notice is that the latency decreases a hop 7 which at first did not make any sense until I thought about it. The only explaination I could think of is that UF has a faster return path to Cox than Cox has to it. So when router at hop 6 sends a packet to me, it probably has to go through the previous 5 routers, but when router 7 from UF sends a reply to me, it probably skips some of these Cox routers. The overall effect is a drop in latency when connecting to UF because the return path is much shorter. I may be wrong since this is just speculation based on observations.