"Note that if you're relying on BGP, there are several known conditions where BGP will not converge--ever--and will experience serious churn. This is bound to happen with protocols that manage multiple metrics; all such protocols will be multi-stable in some way or another. During these "non-convergence events," you might well experience both dropping and looping traffic."
I suspect I missed this very interesting question. SAV table is built from the real forwarding path in the data plane. If BGP does not converge, SAV table built based on BGP routing will also have problem. We cannot build an independent SAV table without routing. I am still thinking about the design philosophy behind this... Dan -----邮件原件----- 发件人: [email protected] <[email protected]> 代表 [email protected] 发送时间: 2022年5月6日 20:23 收件人: 'Xingang' <[email protected]>; 'Lancheng Qin' <[email protected]> 抄送: [email protected]; [email protected] 主题: Re: [savnet] Regarding reusing existing routing protocols for SAV//RE: SAVNET WG charter > "When the next hop to a destination node changes, the node will > generate a triggered notification message. The triggered notification > message will update SAV tables and SAV graphs in nodes along the new > path. IMHO -- if you're going to put more information into BGP, do so as a new message type, rather than as yet another AF ... BGP is already heavily overloaded. Anything that adds to the churn of updates/etc. in BGP could potentially degrade the entire routing system. If you're using BGP merely as transport, add a new message type so the code and functionality can be largely separated from existing BGP mechanisms. > " Just like packet loss from temporal loop during routing update > cannot be completely avoided." In link-state protocols this is true. In path-vector protocols (like BGP) and distance-vector protocols, you generally end up temporarily dropping traffic rather than looping it during convergence events. Note that if you're relying on BGP, there are several known conditions where BGP will not converge--ever--and will experience serious churn. This is bound to happen with protocols that manage multiple metrics; all such protocols will be multi-stable in some way or another. During these "non-convergence events," you might well experience both dropping and looping traffic. /r -- savnet mailing list [email protected] https://www.ietf.org/mailman/listinfo/savnet _______________________________________________ rtgwg mailing list [email protected] https://www.ietf.org/mailman/listinfo/rtgwg
