Yes, peering with the BGP Route Servers is available free of charge to all participants, as one of the many value added services at the INXes.
If you have a simple peering policy, and you're happy to peer with everyone, then there's no doubt about it; you should definitely be peering with the BGP route servers. Peering with the route servers will allow you to be able to exchange traffic from other like minded peers, without going through the hassles of setting up individual bilateral sessions. For example, there are many CDN and DNS operators at the JINX, and having a BGP session to the route servers will help to ensure that you make use of their services seamlessly.
Yes, this is true. This is one of the reasons why larger networks prefer not to peer with the route servers. At the INXes, we provide you with an easy to use, BGP community based, policy tool, where you are able to choose, how to manage your routing policies. More details about this, are available on the BGP Route Server page.
That said, peering with the route server is not something that is dangerous either. We are particularly careful about how configuration is generated, and, we filter strictly and probably better than most networks do.
Do both No, really, we mean it!
A bilateral sessions means that, if there is ever a problem with the BGP-RS, you'll be unaffected. Peering with the BGP-RS means that, until you get around to setting up bilateral sessions, you'll get the benefit of new peers that come onboard, and are already peering with the BGP-RS.
If this (peering with BGP route servers) is not considered best practice, why do you encourage it?
It's our belief that network engineers should learn and build their capacity to run their networks, so we strongly encourage you to setup bi-laterial sessions to peers that are of particular importance to you. At the same time, we recognise that operations teams are always busy and aren't able to get around to these tasks as quickly as might be possible. Multilateral peering (ie. via the route servers) will give you the benefit of being able to pickup traffic for networks you may not have direct sessions with immediately, and bilateral sessions will give you the piece of mind that if something ever happens to the IXP route servers, your peering will continue uninterrupted.
It's a relatively simple way to see which networks, and which prefixes are available at the IXP via your peering sessions. We also provide simple diagnostic tools (ping, traceroute) for network debugging purposes. A web front end to each of the route collectors operated by INX is available via https://lg.inx.net.za
The collector (and its interface via the looking glass) act as a neutral view into multiple networks. By peering with the collector, you're allowing your peers to perform simple debugging techniques (like ping and traceroute) that can aid them in their work. In the same manner, third party networks that do this, give you the same benefit, for your network.
The route collector also provides an independent tool, to ensure that what you think you are advertising to your peers, is actually what your peers are receiving! And since it takes a few minutes to setup, there's literally no down-side to this.
Remember that a big part of JINX is community And JINX is old. Different research groups often collect data in different ways, and we are impartial to that. We encourage and support all research.
The INX route collector does not advertise any prefixes back to you. Some other route collectors might advertise a single, unused prefix, simply to let you know that the session is working as intended.
Since the route collector does not advertise any prefixes back to its peers, there is no risk in peering with it. The policy of "non-advertisements" has been hard-coded into the BGP collector setup in at least two places, so there is no chance of this breaking. It is simply not possible to hairpin peering traffic through the route collector.
That's quite fine; we'd still welcome your peering session to the route collector, since peering with the route collector is not an indication of network policy, nor does it allow other participants to abuse your routing policy.
In general you want as much traffic to go across your peering interfaces as possible. But blindly setting your BGP local preference to a random high number without thinking through a complete policy, is a Really Bad Idea. It's a much better idea to plan out a BGP local preference strategy for your network, and then execute according to that. Here are some ideas to help you get started below, keeping in mind the general principle that localpref works on the basis of client_localpref should be higher than peering_localpref which should be higher than transit_localpref. Substitute the numbers for ones that make sense in your network
|Customer prefix||Primary prefix learnt from customer||999|
|Customer backup||Client backup prefix||950|
|Domestic peer||Peers within a metro||899|
|Domestic peer backup||Backup for metro peering||850|
|Peers from a defined region||799|
|Regional peer backup||Backup for regional peers||750|
This is just a rough idea of system that could work. You're encouraged to spend time to think about where your entry and exit points are, how you would like traffic to flow through your network, and how to design for efficiency.
There are many forums where you can get assistance with BGP. We recommend looking through some of the BGP course materials from the African Network Operators' Group training material, and joining a local technical community and asking for assistance there. Additionally, we perform free BGP and network design training classes periodically, so feel free to contact us for assistance.