nanog mailing list archives

Re: Peering Latency


From: Ca By <cb.list6 () gmail com>
Date: Thu, 3 Jul 2014 09:52:29 -0700

On Jul 3, 2014 9:47 AM, "Sam Norris" <Sam () sandiegobroadband com> wrote:

Hey all - new to the list but not to the community...

Wondering if this is typical when there is too small of a pipe between
peering
arrangements:

From Level3 to Time Warner

     ADDRESS                                    STATUS
   2    4.69.133.206 4ms 4ms 4ms
   3    4.69.153.222 9ms 4ms 4ms
   4     4.69.158.78 8ms 4ms 4ms  (L3)
   5    66.109.9.121 28ms 53ms 29ms   (TWC)   <------
   6    107.14.19.87 30ms 28ms 28ms
   7    66.109.6.213 27ms 28ms 28ms
   8      72.129.1.1 32ms 32ms 32ms
   9      72.129.1.7 27ms 26ms 25ms
  10   67.52.158.145 28ms 29ms 31ms

From TWC to Level3

 # ADDRESS                                 RT1   RT2   RT3   STATUS

2 24.43.183.34                            5ms   5ms   6ms
 3 72.129.1.14                             8ms   8ms   8ms

 4 72.129.1.2                              6ms   8ms   8ms

 5 107.14.19.30                            7ms   8ms   8ms

 6 66.109.6.4                              8ms   8ms   8ms

 7 107.14.19.86                            5ms   5ms   5ms

 8 66.109.9.122                            34ms  33ms  31ms  (TWC)
 <------

 9 4.69.158.65                             31ms  30ms  29ms  (L3)
10 4.69.153.221                            33ms  33ms  34ms
11 4.69.133.205                            32ms  32ms  31ms


I am showing, typically at night, a 20-40ms jump when hopping from Level3
to
Time Warner and back in Tustin, CA.  This does not happen when using
Cogent or
other blended providers bandwidth.   I believe they are probably stuffing
too
many bits thru the peering there and wondering whats the best way to
prove to
them both (we pay for both) that they need to fix it.

During non-peak traffic times these look normal (sub 10s).

Sam


This latency usually means a change in the return path as you cross an AS
boundry.

The first AS may have a local peering for the best return path while the
2nd AS on the return path has to go to a different region to take the bgp
best path


Current thread: