On Mon, Aug 9, 2021 at 4:04 PM Robert Raszuk <robert@xxxxxxxxxx> wrote:
Hi Warren,There are dire proclamations that IPv6 TCP anycast cannot work....Maybe I missed it but I do not recall anyone said this here.What have been said is that due to hashing or routing changes you may end up in different endpoint of the same IP address during TCP session lifetime.Now two things needs to be observed:A) That endpoint may not be ultimate dst server but LB, application gateway or smart proxy fully aware about such events and automagically deal with session recreation and transparent to the user re-authentication such that eyeball never sees a glitch.B) Anycast for company X != anycast for company ZTake Google - Company x - they have such great coverage and presence in all most important IXPs that path from any user to their entrance is just few hops away. That means that routing changes which can impact the anycast flows are near zero. Same for different hashing ... worse what can happen to you you will end up on different port of IXP edge router. That all means that in such case anycast is here for user convenience and will work just fine.In contrast take Company B which will inject same /24 from SF, NY, FRA, CDG & SIN ... First imagine FRA & CDG - if any of those happen to be served by same ISP then it is already pretty random where the packet will end up. Such ISP single link flap may switch to different BGP best path and take opposite exit.Bottom line is that when you build a solution consideration needs to be done at all levels from kernel app runs on via customer base and ISPs/IXPs involved in plumbing the packets.
Indeed. I just wish you'd posted this last sentence right when the thread started; it would have prevented the needless annihilation of many electrons...
W
Best,R.
The computing scientist’s main challenge is not to get confused by the
complexities of his own making.
-- E. W. Dijkstra
complexities of his own making.
-- E. W. Dijkstra
