How many BGP peers are supported X670-48x?

  • 0
  • 1
  • Question
  • Updated 3 years ago
  • Answered
I wonder how many BGP peers can I connect to the switch? I have a BGP session to datacenter provider who gives as the default gateway for IPv4 and IPv6 via 2 separate BGP sessions. I have a whole bunch of loadbalancers (>30 machines) which I want to connect to the switch via BGP Each one needs to establish 2 sessions for IPv4 and IPv6 (restriction of BIRD).

How many BGP sessions are safe for this switch? I'm currently using OSPF for this purpose but I'd like to give BGP a try for some reasons.
Photo of Kajetan Staszkiewicz

Posted 3 years ago

  • 0
  • 1
Photo of welisson

welisson

  • 908 Points 500 badge 2x thumb
Hi Kajetan,

I belive they don't have any limitation about amount BGP Peer, but so, about amount routes received and installed at RIB.
However i advice you about using Peer Group, is more simple and easier to configured also administration.
Photo of Grosjean, Stephane

Grosjean, Stephane, Employee

  • 12,552 Points 10k badge 2x thumb
Hi,

as per the Limits section in the RN, it's up to 128 BGP Peers with default timers.
Photo of Zelnosky, Kevin

Zelnosky, Kevin, Employee

  • 1,124 Points 1k badge 2x thumb
Kajetan,

What version of EXOS are you using?  The supported limits can be found in each of the release notes PDFs.
Photo of Chad Smith

Chad Smith, Senior Escalation Support Engineer

  • 5,620 Points 5k badge 2x thumb
Kajetan,

As Stephane has said, BGP peer limit is 128.  However, it is important to determine how many routes will be learned from each peer as it is more likely you will reach the route table limitations when adding a large number of peers.

X670 maximum Unique IPv4 BGP Routes: 25,000
X670 Maximum non-Unique IPv4 BGP routes: 25,000
X670 Maximum Unique IPv6 BGP Routes: 8000
X670 Maximum non-Unique IPv6 BGP Routes: 24,000

It is also important to understand that the routing table limits are shared.  These numbers are maximums, assuming no other routing protocol is in use.  Using multiple routing protocols or using IPv4 and IPv6 will reduce the maximum limits for each protocol.
Photo of welisson

welisson

  • 908 Points 500 badge 2x thumb
Chad when you said unique IPV4, did you mean is it the amount route that x670 support in FIB and non-unique will be RIB?
Photo of Chad Smith

Chad Smith, Senior Escalation Support Engineer

  • 5,620 Points 5k badge 2x thumb
Yes that is essentially what it means.  For example, a non-unique route would be a redundant route learned from another BGP peer that is not preferred/active. 
Photo of welisson

welisson

  • 908 Points 500 badge 2x thumb
thank you Chad,
in this case when the switch received the routers from bgp peer, and it is exceed limitation non-unique and unique, what is the behavior expected of routes exceeded?
In this case it will fill at RIB just amount permit on RIB, or will it received the full-routing and take crashed the switch?
Photo of Chad Smith

Chad Smith, Senior Escalation Support Engineer

  • 5,620 Points 5k badge 2x thumb
Generally speaking, you should not exceed the supported limits.  Once the hardware routing table is filled, traffic from all other routes would have to be slow path forwarded by the CPU.  This is not ideal.  Eventually, if the limits were sufficiently exceeded it is possible a crash may occur.  

The following GTACKnowledge Article has some information that you may find interesting regarding this topic:  Unexpected Reboot When BGP Learnt Routes Exceed the Supported Limit.
Thanks for pointing me to Release Notes. I've checked only the usual huge manual PDF file for firmware 15.5 which I'm running in the test enviroment and I could not find this information there.

I want to run quite short timers. Currently we use on-link public IPs on Loadbalancers provided by FreeBSD's CARP mechanism which detects trouble in just (AFAIR) 3 seconds. So let's say that I want BGP keepalive every 3 seconds and and hold-down after 10 seconds.

Number of routes should not be an issue. Each Loadbalancer will advertise only his single IPv4  /24 prefix  and a single IPv6 /64 and a few Loadbalancers will advertise some longer prefixes during things like migration of services or for redundancy between datacenters. The only thing received from Datacenter is default gateway.

Of course I want to use peer groups. But does it change anything in terms of maximum number of peers?
Photo of Frank

Frank

  • 3,662 Points 3k badge 2x thumb
Kajetan,

With this many bgp peers, would it perhaps make sense to run a "Route Server" or "Route Reflector"? Basically a central BIRD/Quagga server that everyone peers with? (See http://www.cisco.com/c/en/us/td/docs/ios/ios_xe/iproute_bgp/configuration/guide/2_xe/irg_xe_book/irg... and http://www.gossamer-threads.com/lists/quagga/users/11604)

Note: I may be totally off here; I'm just currently starting to read up on those things and it seemed as if it might be useful in this case. But as I said, I might be totally wrong ;)
Photo of Chad Smith

Chad Smith, Senior Escalation Support Engineer

  • 5,620 Points 5k badge 2x thumb
Kajetan,

Changing the timers increases the processing load on the switch.  However, you also have a small number of routes.  Maximum limits are usually best case scenarios.  With the mixing and matching of IPv4 and IPv6 and changing the keep-alive it is difficult to determine your max. I would have to recommend you test it just to be on the safe side. 

You are limited to a maximum of 64 BGP peer groups. I do not think that this will have an impact in your setup unless you need more groups than that.
Thank you everybody for replies,

At this moment with a total of 14 peers bgp process shows nearly no load (<1%,  with 3,5% peak). As for external route-reflector setup, I want to avoid it in order to keep the amount of hardware needed for network operation to minimum, that is fit as much as possible on central Extreme switches, which are hardly ever restarted and don't need as much maintenance as "software" bgp server.