Dear Juan Brenes, Sun, Oct 22, 2017 at 12:54:27PM +0200, Mirjam Kuehne wrote: > Please find this new article on RIPE Labs by Juan Brenes, one of the > RACI fellows at RIPE 75:
Thank you for your effort to explore possible BGP improvements. Seeing that you are part of the RACI program, and likely to be a newcomer to the RIPE community, I want to emphasize that the feedback offered in this email is only meant to help improve the proposal and for the benefit of all. Welcome to RIPE! > Power Prefixes Prioritisation for Smarter BGP Reconvergence > > Every day the Internet experiences failures that affect the > reachability of certain zones (i.e prefixes), such as cities, > countries, etc. Whenever such events happen, the routing information, > which is the path that information must follow in order to get from > one place to another, must be updated so that the failure is > circumvented and the zone becomes reachable again. > > https://labs.ripe.net/Members/juan_brenes/power-prefixes-prioritisation-for-smarter-bgp-reconvergence It appears an analysis was done on a BGP implementation which suffers from the concept of prefix _dependent_ convergence (PDC), in context of PDC the convergence time grows dependent upon how many prefixes are in your FIB/Loc-RIB. Within that context I can envision scenarios where a ranking of prefixes can be beneficial if traffic volume is used to order the work to be done. However, the industry has seen quite some innovation in recent years. One of the foundational concepts for blazing fast convergence is "BGP Prefix Independent Convergence". The full specification can be reviewed here: https://tools.ietf.org/html/draft-ietf-rtgwg-bgp-pic-05 I see from your article that the Quagga software was used in your research, as far as I know BGP-PIC was never implemented for Quagga. Another crucial concept for fast convergence is the availability of backup paths, one method is to use ADD-PATH (RFC 7911). By combining ADD-PATH and BGP-PIC near instant convergence can be achieved. With the above in mind, I have some questions: 1) Would it perhaps make sense to rank updates by BGP next-hop (using a metric like traffic volume), rather than destination prefixes themselves? Each BGP next-hop in effect represents a group of prefixes. 2) Is the BGP-PPP ranking still useful when BGP-PIC and ADD-PATH are used on both sides of the BGP session? 3) Are you looking to publish your work in context of IETF or IRTF? > Juan will present this topic at the Routing WG session later this week. I look forward to your presentation! Kind regards, Job
