Level(3) and IPv6 Day

[Bernhard Schmidt]

On Wed, Jun 08, 2011 at 11:29:24AM -0400, Paul Stewart wrote:

Hi,

> WOW... we have IPv6 from Level(3) and I thought you were joking a bit until
> I did a traceroute that leaves Toronto (on a different provider) and goes
> through Europe (Paris etc) and then comes back to then go west of us into
> Denver:

Every provider is responsible for their own outbound paths. Looks like
Tiscali fixed it quickly this time...

 2  2001:1608:0:11::1 (2001:1608:0:11::1)  75.038 ms
 3  ae0-233.ip6.tinet.net (2001:668:0:3::2000:301)  0.961 ms
 4  2001:1900:5:3::51 (2001:1900:5:3::51)  0.941 ms
 5  vl-79.edge3.Frankfurt1.Level3.net (2001:1900:104:2::8)  8.065 ms
 6  vl-4060.edge4.Paris1.Level3.net (2001:1900:5:1::216)  16.385 ms
 7  vl-4081.edge3.Paris1.Level3.net (2001:1900:5:1::12d)  17.096 ms
 8  vl-4086.car1.Washington1.Level3.net (2001:1900:6:1::16)  91.781 ms
 9  vl-4083.car2.Washington1.Level3.net (2001:1900:4:1::de)  254.125 ms
10  vl-4061.car1.NewYork2.Level3.net (2001:1900:4:1::106)  92.461 ms
11  vl-4080.car2.NewYork2.Level3.net (2001:1900:4:1::f2)  91.437 ms
12  vl-4061.car1.Chicago1.Level3.net (2001:1900:4:1::21)  117.377 ms
13  *
14  vl-4041.car2.Denver1.Level3.net (2001:1900:4:1::35)  154.301 ms
15  vl-4080.car1.Denver1.Level3.net (2001:1900:4:1::2d)  140.442 ms
16  Level3-MOSS.vl-956.car1.Denver1.Level3.net (2001:1900:4:2::fa)
141.162 ms

This is unfortunately not rare with Tiscali. For years I have frequently
seen bad paths with them, in 99% of the cases through IIJ, to
destination networks where I'm sure they have better (even direct) paths
to. I'm not exactly sure what they are doing, from the outside it looks
like a mess. 

For this occasion I have reused a very ugly script I wrote for RIPE56 in
Berlin to have a look at the various paths between major ISPs from the
viewpoint of the BGP routeserver at grh.sixxs.net. Note that "major" is
my personal definition, I'm not going to elaborate why I omitted or
forgot one ASN or the other.

              174     1239     1273     1299     2914     3257     3356     3549     6453     6939
     174:              28/-     21/-    318/-    473/-     47/-    328/-     91/-    125/-      -/-
    1239:     98/-               -/-    169/-      -/-      -/-    166/-    171/-     99/-      -/-
    1273:    197/1     -/15             211/-    491/-    393/-    383/-    404/-    242/-   2241/-
    1299:    319/-     50/-     32/-             584/-    406/-    730/-    227/1    273/-   1688/1
    2914:    272/3     51/-     35/-    270/-             296/1    377/-    479/-    174/-   2084/1
    3257:    329/1     -/16     20/-    341/1   463/54             549/-    389/1    136/1   1789/4
    3356:    299/4     88/-     41/-    581/-    664/1    592/2             979/-    285/1     -/11
    3549:    187/1     43/-     23/-    153/1    431/-    461/1    416/-             247/-   1565/3
    6453:    245/1     49/-     29/-    261/-    435/-    220/-    388/-    501/-               2/-
    6939:      -/-     44/-     17/-    209/-    249/-    156/1      -/-    270/-    160/-         

(Table for the MUA-impaired also on http://pastebin.ca/2076834)

It shows the number of prefixes with BGP paths from Source ASN (down) to
Destination ASN (right). The first number is a direct path (has a A_B
edge), the second number is an indirect path (has an A_!B_B path).

First thing to notice, there could be connectivity issues between the
following players:

A/B			A->B		B->A
174/6939:		-/-		-/-
1239/1273:		-/-		-/15		1)
1239/2914:		-/-		51/-		*)
1239/3257:		-/-		-/16		2)
1239/6939:		-/-		44/-		*)
3356/6939:		-/11		-/-		3)

*) I think those are false positives due to lack of view from 1239
downstreams.

When you look closer at the indirect paths, you get

1) 1273->1239:
	1273 6175 1239 1239 1239 1239 1239 1239 1239
	1273 1257 1880 1239

2) 3257->1239:
	3257 2497 6175 1239 1239 1239 1239 1239 1239 1239
	3257 1257 1880 1239
	3257 2497 701 12702 286 1239

Here we have the crack paths through IIJ (2497)

3) 3356->6939:
	3356 3257 7473 24203 6453 6939
	3356 32748 6939
	3356 3257 10026 17888 4635 6939
	3356 3549 10026 17888 4635 6939
	3356 23308 6939
	3356 8057 6939
	3356 35819 6939
	3356 5400 8903 6939
	3356 20811 6939

Wow, Level3 seems to have some customer filtering issues.

But the most interesting point is this in my opinion

2914->3257:	296/1
3257->2914:	463/54

So, there obviously is a direct peering between them, but still Tiscali
sends 54 prefixes through other networks to get there. And, guess what
the paths are

	3257 2497 2914
	3257 2497 701 12702 286 2914

Best Regards,
Bernhard