Re: [Idr] IDR WG LC on draft-ietf-idr-ix-bgp-route-server (2/2/2015 - 2/16/2015)

[Andy Davidson <[e-mail]>]

On 2 Feb 2015, at 17:36, Susan Hares <[e-mail]> wrote:

> WG LC  is from 2/2/2015 to 2/16/2015
[…]
> This begins a 2 week WG LC for draft-ietf-ix-bgp-route-server-06, and a 2 week WG adoption/LC for the associated WG report.
> Please comment on if you feel this is useful for operating exchange points, and include in your comments support or no support.

Colleagues,

Route-servers are an operational reality and norm at modern Internet Exchange Points.  80% of exchange points belonging to the European Internet Exchange Association (Euro-IX) operate a route-server service for their exchange participants [0], and in my anecdotal experience as an operator of two other exchanges in the UK (LONAP - 100Gbit/traffic, 150 participants & IXLeeds - 4Gbit/traffic, 15 participants), they are requested, expected, and **USED** by over three quarters of exchange participants.

Route-servers are growing in popularity, in 2010 only half of the Euro-IX membership operated and offered route-servers.

This seems to be a worldwide phenomenon, I also work with the NapAfrica exchange which has a 94% adoption rate for the route-servers on the largest peering LAN and a 100% adoption rate on the smallest peering LAN.  They’re also a popular service on exchange points I help run in North America.  Several exchange point operators in Asia have previously asked for my route-server configuration generator scripts (I pray that are all migrated to IXP Manager now).

A great deal of work has been undertaken by the Euro-IX and wider peering community to make route-servers more reliable.  Back in the dinosaur days of the late 2000s, we didn’t have route-servers that could support multiple-RIBs, and had to put up with route-servers with poor convergence performance leading to blackholing of traffic whilst withdraws trickled through poorly performing route-server code. We started to write down what we wanted, no needed, the route servers to DO, and worked with the software vendors to get these features built in.  The result was better performing, tested software that was compliant against this stack.  [1]

The idea that our loosely defined, and these days only committed to memory, list of wishes could instead be a document … an IETF document! … with wider industry support, that can be used to describe our needs to the next generation of route-server software vendors is fantastic.  I fully and wholeheartedly support the publication of this document.

We’ve accidentally, in this thread, gone down the path of considering whether they are the best way forward for multilateral/mass peering.  No they’re not, and let’s do some work to find something better.  But MLPA route-servers are here for a long time to come, so let’s publish this excellent specification document right away.





[0] https://www.euro-ix.net/documents/1467-euro-ix-route-servers-stats-pdf?download=yes

[1] http://www.uknof.org.uk/uknof15/Davidson-Bakeoff.pdf



— 
Regards, Andy Davidson

Relavent hats to this conversation:
Board member, European Internet Exchange Association
Director & Chairman, LONAP Ltd., IXLeeds Ltd. (United Kingdom)
Council member, NapAfrica (South Africa)
Vice President European Ops, IIX Inc. (North America)