On 1/21/22 8:22 AM, George Ross wrote:
Would it help if you announce different MED values? That way upstream should prefer one while it's up, but fail over to the other when it's not.
I'd be inclined to investigate different MED values too. Though I'd probably announce two /24s, which would allow you to have crossed MED values. E.g. router A has one MED for the first /24 and a second MED for the second /24 while router B has the second MED for the first /24 and the first MED for the second /24. The hope is that each /24 will have affinity to one router, thus pseudo load balancing and utilization of both links, while still being an active backup for the other /24. If your problem is an out of order / reordering issue, I'd expect that sniffing traffic on any system in the /23 would make this very apparent in Wireshark. TCP /should/ handle this without any problems. Are you by chance doing any SPI on either of the routers? Is there any chance that the timeout is actually related to traffic flows switching which router they are coming in on and the new router's SPI not having state, thus evicting the flow? -- Grant. . . . unix || die