[j-nsp] Full BGP table, one provider w/ 2 routers, slow forwarding convergence

[Clarke Morledge]

I wanted to follow-up on the list with a report of what I did to try to 
resolve this issue.

I have not had the opportunity to try to play tricks with routing policy 
to only populate the forwarding table with a default router while still 
receiving a full Internet router table via BGP.  It is a little bit tricky 
for me since I also have another ISP to worry about that complicates our 
BGP design.

However, what I did try was the load balancing method across both of my 
outgoing links using the "multipath" option described here:

http://www.juniper.net/techpubs/en_US/junos13.3/topics/topic-map/bgp-multipath-unequal.html

It breaks my plan to do active/passive between the two links for outgoing 
traffic. I am using MED to control preference for incoming traffic, which 
works well for an active/passive design for incoming traffic.

I was able to reduce the failover time in the event of a "fiber cut" type 
of issue from 90 seconds down to about 10 or 20 seconds for most traffic. 
However, this only works if you are receiving the exact same copy of the 
Internet routing table from the upstream provider.

In my case, there is enough variation in routes coming from the upstream 
provider between their two routers that I do not really get full load 
balancing going on in the forwarding table.  In those cases, I am back to 
about 70 to 90 seconds to complete a failover transition for outgoing 
traffic.

I did upgrade to 13.3R3 to try to resolve PR963060, but I am still seeing 
enough traffic loss that it makes me skeptical as to whether I have 
really hit this PR, if the issue has not really been resolved, or if 
something else is going on.   I still think Junos should converge faster.


Clarke Morledge
College of William and Mary