Route advertisement with AS path

[viswa sai]

Hello

We are running Multi-homed network, to influence the BGP route selection, we are using AS path attribute with route-maps.

Recently, we observed that the routes advertised on TCL network(ISP 1) were not reflected in global routing table.

For example, we have advertised a network (196.X.X.X/24) on TCL BGP peer as best path and also advertised the same network on another BGP peer with AS path pretend of 10times. But, the network is reaching via Bharti BGP (iSP 2 )peer instead of TCL peer.

Can anyone help understand as why it is preferred route is via Bharti 

Thanks

Viswa Sai

[Cisco Freak]

Hi Viswa,

Can you please share your configs?

CF

[viswa sai]

I have resloved this issue by advertising the /19 perfix with a single as prepend on both ISP's

Now, it seems working, when i shut one peer the traffic is switching to another ISP. but the catch is sometimes it takes more time to converge the traffic..

any idea?

[Ravindra Simpi]

Network statement in BGP configuration is used to identify which networks are being advertised. BGP process then checks the global routing table, if it sees a prefix in global routing table and with exact match (including subnet mask), only then it will advertise that network to other BGP peers. 

Is this network a local network or learned from other routing protocols? If locak, make sure you enter exact mask of the network seen in routing table. If learned from other routing protocols, the better way is to selectively redistribute iGP routes into BGP using prefix-list and route-map.

As far as convergence is concerned, below is explanation:

1. BGP routers router will not start the BGP Best-Path calculation/selection process until they receives all NLRI from BGP peer. This will be known from UPDATE messages. End of UPDATE messages is usually identified after a KEEPALIVE message is received. 
2. The time taken to learn new best path is directly proportional to number of NLRIs received from peers. 
3. Only when your service provider router selects it's best path, and installs into RIB, it is going to send UPDATE message to your routers. If SP routers use line cards with Cisco distributed forwarding, it is going to populate it's FIB and then send UPDATE message.
4. It depends on how fast your Bharati BGP peer detects your network unreachable and sends UPDATE messages to it's peers to withdraw your network's NLRI from it's routing table.

There are ways to improve this convergence, but at service provider level. In your network, if you want faster re-convergence, static routes (with higher AD) would be a wonderful solution.

Few other ways would be to use:

1. Bidirectional forwarding detection (BFD)
2. fast neighbor failover
3. BGP next hop tracking
4. BGP best external path (IOS and vendor specific)
5. BGP prefix convergence (IOS and vendor specific)

Peace and Health,

Ravindra

[viswa sai]

Hi - Thank you for your detailed information.

During my fail over test to check the prefix advertisement, i noticed that sometimes supernet /19 network was immediately available in global routing table and the other time it took more than 30 minutes. During the checking process i clear the BGP session of both Active and down peers  and then i am able to view my prefix(/19) and then able to receive the reverse traffic on active peer.

So, can a clear ip bgp soft out on the active peer. will that initiate the NLRI update process and sends the BGP update message to  the upstream peer?

I am still finding my feet on the ground and haven't tried the other options i will test them..

Appreciate, if you reply again.

thanks :-)

[Ravindra Simpi]

Yes you are right. 'clear ip bgp soft out' command will send new set of UPDATE to BGP peer without re-configuring the routing table and clearing BGP session.

One question, why are you advertising /19 prefix into BGP session connected to your service provider routers. If /19 is the network you really want to advertise (because you have complete block of IP address in your network) then it is good, else you would create a black hole in your network. Because SP routers will start forwarding traffic for subnets that are not part of your network.

Peace and Health,

Ravindra

[viswa sai]

Thx for clearing my doubt regarding update.

The reason for advertising a /19 is for having a redundancy on ISP peer.

I have a 10 /24 network and for doing load balancing between two bgp peers, i advertise 5 /24 prefix's on one peer and another 5 on another peer.  Earlier in the first post i mentioned the reason..(pl refer). When i advertise  a network (/24) on the another peer with a AS-PATH prepend for network reachable in-case of my primary ISP fails.

Recently, i observed the network started accepting Bharti network as best path to reach in-spite of having AS-PATH prepend (4 or 5 times of that particular network) i observed that they are re-advertised in global routers with higher local pref. So i am unable to acheive the optimal load balancing.

So, i sorted out this issue, by announcing a /19 prefix on both ISP's  so if one of the peers goes down, i need not advertise manually.

Hope i am clear!!!.  is it the correct a way of achieving load balancing

[Ravindra Simpi]

Several options:

Option-1: Splitting your /24 network into two /25 networks. And then advertise /25 network over the link along with original /24 network. But you need to check this with your SP in 1st place.  Most SPs won't allow network advertisements of less than /24 to be advertised to their network.

          You must make sure that you advertise your /25 network with NO_EXPORT community string so that /25 won't leak outside your SPs AS. Or check with your SP what community to be tagged with /25 if it is allowed to be advertised.

Option-2: You may also consider buying a /22 subnet from your SP to achieve the load sharing requirement you are looking for. Buy /22 subnet, split it into two /23 subnets and advertise /23 along with /22 on links to SP.

I personally feel advertising /19 prefix is not a good idea when ISP has provided us just a /24 prefix. We will end up in more problems later.

Peace and Health,

Ravindra

[viswa sai]

Thank you for the inputs.

I have a 32 /24 networks and i am just announcing a supernet block to my upstream.

[Ravindra Simpi]

I understand. 

Just take care that your network won't attract transit traffic for other subnets that are part of /19 but not owned by you.

I am also interested in knowing about the document/white paper which you are referring which explains about advertising a less specific prefix into BGP. Can you share me that info?

Peace and Health,

Ravindra

[viswa sai]

we own all the 32 /24 network so there is no way i can attract the transit traffic.

I recently started facing this issue. whenever i advertise with more AS-PATH prepend of a particular network on other ISP peer for redudancy.

I don't have a white paper but i  have read in NANOG forum regarding this announcing supernet prefix. Its working but as i said earlier, sometimes the convergence is taking time for learning the prefix's under /19 :-(.

[Ravindra Simpi]

Oh okay. I was still considering your previous statement

"I have a 10 /24 network and for doing load balancing between two bgp peers, i advertise 5 /24 prefix's on one peer and another 5 on another peer."

Then you can split your /19 aggregate into two /20 networks. And then advertise /20 network over the primary link along with original /19 aggregate. 

Peace and Health,

Ravindra