[j-nsp] Convergence time

[Dovid Bender]

Hi,

We currently have two MX5's with three upstreams (two of which have a BGP
session with each MX5). We wanted to get full routes from all upsteams but
were told by a few people at the time that should one peer drop out it
would take 2-3 minutes before the MX5 would pull the routes for the
upstream from the routing table. Now I am being told that it's not correct
and should I lose a peer any routes that we got from that peer would
instantly be dropped from routing table and the traffic would go to any
other upstream that gave me that route or my static route. Is there any
truth to this?

TIA.

Dovid
_______________________________________________
juniper-nsp mailing list juniper-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/juniper-nsp

[Rolf Hanßen]

Hi,

times will depend on the type of failure.
If the upstream router fails but your link does not go down (for example
if there is a switch between the routers), you will have a BGP session
timeout after 3 minutes (with default config) and then the router starts
to change the routes, which can take even longer on such slow hardware.

Could be that this helps at least in case the ports goes down:
https://www.juniper.net/documentation/en_US/junos/topics/task/configuration/bgp-configuring-bgp-pic-for-inet.html

Could also kill the box because of increased memory usage.
But don't you have memory issues anyway if you run full tables on an MX5?

He did some benchmarking with MX104:
http://seclists.org/nanog/2018/May/470
Maybe just add a bit because MX104 is a bit less slow than MX80 boxes.

kind regards
Rolf

Post by Dovid Bender
Hi,
We currently have two MX5's with three upstreams (two of which have a BGP
session with each MX5). We wanted to get full routes from all upsteams but
were told by a few people at the time that should one peer drop out it
would take 2-3 minutes before the MX5 would pull the routes for the
upstream from the routing table. Now I am being told that it's not correct
and should I lose a peer any routes that we got from that peer would
instantly be dropped from routing table and the traffic would go to any
other upstream that gave me that route or my static route. Is there any
truth to this?
TIA.
Dovid
_______________________________________________
https://puck.nether.net/mailman/listinfo/juniper-nsp

_______________________________________________
juniper-nsp mailing list juniper-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/juniper-nsp

[Melchior Aelmans]

Hi Dovid,

MX5 might not be the ideal platform handling multiple full table BGP
sessions. This is due to it's somewhat slower CPU compared to other MX
routers.
You might want to limit the amount of full table or flapping sessions or
get yourself another platform. For example MX150 or 204 wouldn't have any
problem converging this amount of full tables.
Does this help?

Regards,
Melchior

Post by Dovid Bender
Hi,
We currently have two MX5's with three upstreams (two of which have a BGP
session with each MX5). We wanted to get full routes from all upsteams but
were told by a few people at the time that should one peer drop out it
would take 2-3 minutes before the MX5 would pull the routes for the
upstream from the routing table. Now I am being told that it's not correct
and should I lose a peer any routes that we got from that peer would
instantly be dropped from routing table and the traffic would go to any
other upstream that gave me that route or my static route. Is there any
truth to this?
TIA.
Dovid
_______________________________________________
https://puck.nether.net/mailman/listinfo/juniper-nsp

_______________________________________________
juniper-nsp mailing list juniper-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/juniper-nsp

[Gert Doering]

This post might be inappropriate. Click to display it.

[Raphael Maunier]

MX80 (5-10-40) is a very old platform with a cpu less powerful than your cell phone.
2-3 minutes seems to be a very good performance for this platform :)

I had a customer with 2* MX80 with peering and 4 full feed (95% or ram used !), and I have seen more than 8 minutes during a flap. Convergence is a pain on this hw.
Unless you really need to have the full feed, you should consider to get a default route from your upstream and filter greater than /22 (or 21).

MX204 will be the best choice (mx150 is a cpu based platform)

Raphael



Envoyé de mon iPhone

Post by Dovid Bender
Hi,
We currently have two MX5's with three upstreams (two of which have a BGP
session with each MX5). We wanted to get full routes from all upsteams but
were told by a few people at the time that should one peer drop out it
would take 2-3 minutes before the MX5 would pull the routes for the
upstream from the routing table. Now I am being told that it's not correct
and should I lose a peer any routes that we got from that peer would
instantly be dropped from routing table and the traffic would go to any
other upstream that gave me that route or my static route. Is there any
truth to this?
TIA.
Dovid
_______________________________________________
https://puck.nether.net/mailman/listinfo/juniper-nsp

_______________________________________________
juniper-nsp mailing list juniper-***@puck.nether.net
https://puck.nether.n