Little Clarification Needed with BGP section of ORD

[Deepak Arora]

1.

ORD Page 244 (Hard Copy)

Under Section Route Reflectors:

"As a side note, if Router B were to reflect the best path back to its clients, the clients would receive redundant information. The issue here is not the redundant information that the clients would receive but the processing that the route reflector requires. In other words, it is recommended to have a cluster with a full mesh of clients if clients are present in a significant number or if the physical topology dictates this to be so. The number of clients needed to consider using a full mesh between them depends heavily on the type of router being used as a route reflector and any other functions it might be performing. In other words, to determine the ideal point for your network, monitor the performance and utilization of the route reflectors at the times it needs to send updates to the clients. Your comfort level dictates when to mesh the clients, divide the load between reflectors, or simply dedicate a router just for the route reflector function."

Can someone help with more explanation on this section.

-------------------------------------------------------------------------------------------

2.

Also in the end of BGP chapter author describes the Route Dampening in a case study but there is nothing described from Design standpoint. I remember reading some where in past about Route Dampening doing more bad than good (Perhaps some Nanog presentation).

Whats your opinion on that ?

Do you guys use BGP route dampening in real world deployments ?

if So, you prefer default values or fine tune those to something else ? 

---------------------------------------------------------------------------------------------

3.

How many chances are there to become transit AS in todays world ? where most upstream ISPs put tight route filter on inbound and outbound updates

---------------------------------------------------------------------------------------------

4. Although " ORF " seems to be a nice feature, How often you guys have seen or implemented it in ref to Point 3 ?

Regards,
Deepak Arora

Evil CCIE

http://deepakarora1984.blogspot.com

[Orhan Ergun]

Hi Deepak , 

See the answers inline. 

Orhan Ergun

CCIE & CCDE

Group Manager

27 Aralık 2014 Cumartesi 23:33:07 UTC+3 tarihinde Deepak Arora yazdı:

1.

ORD Page 244 (Hard Copy)

Under Section Route Reflectors:

"As a side note, if Router B were to reflect the best path back to its clients, the clients would receive redundant information. The issue here is not the redundant information that the clients would receive but the processing that the route reflector requires. In other words, it is recommended to have a cluster with a full mesh of clients if clients are present in a significant number or if the physical topology dictates this to be so. The number of clients needed to consider using a full mesh between them depends heavily on the type of router being used as a route reflector and any other functions it might be performing. In other words, to determine the ideal point for your network, monitor the performance and utilization of the route reflectors at the times it needs to send updates to the clients. Your comfort level dictates when to mesh the clients, divide the load between reflectors, or simply dedicate a router just for the route reflector function."

Can someone help with more explanation on this section.

    I didn't read it it was too long , tight schedule sorry and also Russ White is also in the list, he would answer it.  

-------------------------------------------------------------------------------------------

2.

Also in the end of BGP chapter author describes the Route Dampening in a case study but there is nothing described from Design standpoint. I remember reading some where in past about Route Dampening doing more bad than good (Perhaps some Nanog presentation).

Whats your opinion on that ?

Do you guys use BGP route dampening in real world deployments ?

if So, you prefer default values or fine tune those to something else ? 

Check my article about BGP route dampening on packetpushers. You can find it by searhing BGP Path hunting, hope it helps.---------------------------------------------------------------------------------------------

3.

How many chances are there to become transit AS in todays world ? where most upstream ISPs put tight route filter on inbound and outbound updates

If the policy is not implemented through some filtering mechanisms , chance is there but as you said it is implemented. You can use any router as transit router in any routing protocol , this is not limited to BGP , in general you prevent it with summarization , or in eigrp case stub comes as an extra feature , with BGP ; as-path filtering and so on.---------------------------------------------------------------------------------------------

4. Although " ORF " seems to be a nice feature, How often you guys have seen or implemented it in ref to Point 3 ?

It is negotiated during BGP session setup , almost always is used , you shouldn't use soft reconf inbound unless you have a good reason due to memory consideration and manual update procedure. Check also RT constraint for VPN and MP-BGP case. 

[Daniel Dib]

1. I think the point here is that RR is used to reduce the number of peerings, that does not mean that all parts of the network have to peer with the RR only. In the scenario from ORD there is a part of the network that is fully meshed, this could be a PoP and it has been decided that the number of peerings is acceptable to have a higher level of redundancy and faster convergence. However, to reduce the total number of peerings a RR will spread the routes to the rest of the network.

2. I don't see it used commonly. Modern routes can handle route churn better than what we had 10-15 years ago. If I were to deploy it I would deploy Graded Flap Dampening to penalize root DNS prefixes less than an ordinary /24 from random company.

3. More than you would imagine. Many ISPs are are still sloppy with filters if they even have them. It's an administrative nightmare to keep prefix-lists updated and what do you base it on? Most would base it on RIR information but then you rely on that information being up to date if even existant. To keep up with the updates you would have to rely on scripting. There are a few cases every year, intended or not where prefixes leak and attract traffic, such has happened with Youtube and many others in the past. This is also one reason why people deaggregate and pollute the DFZ, because the risk is less with a longer prefix that someone will accidentally or with intention announce your prefixes.

I usually put a filter with ^$ so that I don't advertise something that didn't come from my own network unless I want to be a transit AS of course.

4. I have never seen it in use. In theory it's nice but there's always a trust issue with having another entity having some control over your network even if it's just for a peering going to that entity.

--
You received this message because you are subscribed to the Google Groups "CCDE Group Study" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ccdegroupstud...@googlegroups.com.
To post to this group, send email to ccdegro...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Deepak Arora]

Thanks Orhan & Daniel !!!

[Janesh A]

Hello Orhan

I would like to participate in your classes. Pls let me know how this could be achieved.

Janesh

CCIE- Wireless