[AusNOG] peering questions

[Tony]

Hi all,

I have some questions regarding peering that I'm hoping the helpful folk on the list might be able to assist me with. Our current position might best be summed up in dot points:
- sub-gig transit from two providers
- own AS
- own IP allocation (/20)
- no peering
- renegotiating transit agreements

I know the theories of peering, but not so well versed in the practical side of it. My main question is in regard to how much inbound traffic we might shift to peering (and so by how much we can reduce transit requirements). At this stage I'm focusing on NSW-IX as a good place to start (they have public LG, which helps analysis). I've run nfdump/nfsen on our inbound traffic and stats by AS (% of inbound total) looks something like this (it's hopefully a fixed width table, but who knows how it will format, apologies in advance):

AS    AS_Name  % 
15169  Google     14.4 *
1221   Telstra    13.2 
20940  Akamai     11.5 *
8075   Microsoft  6.1 *
6461   AboveNet   5.9 
16509  Amazon     5.4 *
2764   AAPT       4.6 
15133  Edgecast   2.2 * 
7474   Optus      2.1 
32934  Facebook   1.8 *?
16625  Akamai     1.3 *?
6185   Apple      1.2
4804   Microplex  1.2 
45333  RP Data    1.1 
14618  Amazon     1.1
2906   Netflix    1.0 *

The 2nd entry of Telstra was a bit strange, so I dug into a bit more and found that about two thirds of the traffic from this AS was coming from a few sources (61.9.193.0/24, 23.205.115.0/24, 60.254.143.0/24 & 72.247.223.0/24) which all resolve to essentially being "Akamai".

I can look up the above AS and determine who is on NSW-IX, but as per a recent post to the list, I don't know if that AS is going to be advertising all of the subnets that I am receiving traffic from on the IX and I also don't know that they are going to prefer the IX over transit arrangements (although I hope they are).

So from the above, how much do you think is realistic to be able to expect to get from the IX ?

I've put an * next to the ones I might think that I would receive traffic from the IX instead of transit. I would also hope that Akamai would start to prefer our users to access a cluster connected via IX, rather than the Telstra one that must be seen as "closest" right now. This would also add perhaps another 8-9%.

Adding it all up, I get an optimistic number of perhaps 40-50%.

Is this realistic ? I've spoken to someone else I know in the industry and they've said they get about half their traffic form peering and half from transit. The above numbers I'm seeing tend to suggest this could be true. The very big unknown is what prefixes from the above AS I would get via peering and which would come via transit instead. Do I need to repeat the exercise but instead of sorting inbound traffic by AS, instead sort it by prefixes and then look to see if those prefixes are advertised on the IX ? This of course doesn't guarantee reciprocity, but is about as close as I can get ?

If people would like to share their transit/peering percentages that'd be great (off-list is fine too & I can summarise/average back to list).
 
Thanks for reading and I hope I'm not asking too many noob questions.


regards,
Tony.