Any designing issues in this scenario

[alichangiz86]

Any design issues? I will be pleased if you reply about this network topology and design.

• The topology is consisting of 3 layers: core, aggregation and edge layer.
• Blue, yellow, purple, green and orange colors indicate the fiber links
• Dotted lines are just indicated the border of the layers
• The edge layer routers are PE
• The aggregation layer routers are both P and PE
• Core layer routers are just P routers
• IGP: ISIS (one area, all router L1)
• RR routers: the two central router in the core layer

[balaji.bandi]

the design does not look as expected to be honest, if you produce to this to any one no one can undersand what you trying to achieve with this diagram, the design also not give fully view.

Look at some cisco CVD for reference :

https://www.cisco.com/c/en/us/solutions/design-zone.html#~all-guides

[kim]

Hi guys

I kind of like the diagram, I would however make the ISIS area an L2 instead, may be less of a hassle later on in case integration with other network areas become necessary. And if possible in the physical topology, I would make sure my distribution nodes were all redundantly connected to the core etc. 

Cheers

[MHM Cisco World]

Core/Agg/Acc is for enterprise campus DataCenter 
your topology is for Service Provider Core 
that two different concept. 
first include SW.
second include high end router.

[kim]

Hey

The 3-tier model is a good model for any network type I'd say. In the service provider domain we do divide the networks modules like that, and often use the same terminology.

Very valid design in either case imho. 

The diagram presented and whether it is optimal / sound, depends on geographical distribution, fiber etc. As for the logical part of the design, a bit high level but good.

Cheers

[MHM Cisco World]

in service provider what is function of Core ? function of Agg ? function of Acc?

[kim]

Well the core is in this case the label switching core, often without any bgp etc. Access is the last mile up including the PE, here you have all the different medias comming in from customers, DSL, Fiber etc.. And the distribution could be the "border" out of a pop for instance towards the rest of the network. Something like that maybe

[MHM Cisco World]

in enterprise campus is the Core must be higher end than Acc ? 

[kim]

Well it would definitely make sense to have devices in the dist / core which each handles more bandwidth, since each tier aggregates more bandwidth than the previous. But it isn't only about capacity. In larger networks the core helps to keep cost and complexity down by reducing the number of connections from the dist to the rest of the network. Another important difference is platform features from access with lots of different transport medias and access technologies over distribution where we are sort of bundling all of that and overlaying the transport of that over our underlying network, the core - which then has to be redundant and have enough capacity for customer traffic.

[filopeter]

Consider to configure ISIS as L2 only (all routers L2 only).
Consider to migrate RR functionality to dedicated routers, let the core routers stay as P routers only.

[alichangiz86]

Thank you guys for your replies. as you said it's a HLD for a MAN network. I cannot find any references in design zone to suite my needs. limitations of project in fibering and digging the land leads to the differences between redundant links.

one of my great concerns is the placement of RRs because I don't want to have sub-optimal routing in my design. (In both scenarios, RRs in core layers routers and independent RRs). Any suggestion for the placement of them?

another one is the core layer. in case I would go to peering my core routers to an IXP, all my routers somehow become a PE router (all of them do POP and PUSH) do you think is it OK?

[kim]

Hey

You are right the placement of your RRs is important - and I agree with @filopeter. Your RRs are control plane nodes, and as such don't need to be placed on nodes in the core. It is perfectly okay to place them in separate datacenters for instance as virtual routers where you can more easily size the up as needed.
I would also start thinking about the fact that your RRs simplify your forwarding tables by only presenting the network with only one best route. Is this good for you, for scale for instance, or do you need to have more paths in the network etc.
Also, consider the benefits of splitting your control plane up into address families by placing for instance EVPN away from L3VPNs in different RRs. 

Personally I prefer to keep the core free of PE functionality and simply setup an extra set of PEs for IX'ing and other stuff. I want to keep my mpls core as stable and steady state as possible. As soon as you introduce PE functionality on a core node you also expose that node to the instabilities of config errors, troubleshooting mistakes, qos policies and what not. 
 

[alichangiz86]

Thank you @kim for your actively participating in this topic.

The idea behind all router just L1 is coming from a blog which explained CCIE topics. I think I should change my mind.

I'm going to have Unequal load Sharing in IGP (all routers are IOS-XR) and MPLS traffic engineering so I will have multiple paths to reach to my destinations. As you know RRs present the best routes from their point of views. In my plan, I believe I will have sub-optimal routing issues. So, I decided to utilize BGP Optimal Route Reflection feature in my LLD. Have you ever experienced with this feature? Any experience will be valuable to me.

[filopeter]

Hi,

from my understanding BGP ORR is used to find the best AS exit point (eBGP peering). In iBGP environment, in case of load sharing inside your AS, I would consider to implement iBGP multipath.

[MHM Cisco World]

@kim you are right I found doc. about using tier in SP design 
https://fredrikjj.wordpress.com/2014/07/07/bgp-design-implementation-part-14-service-provider-architecture-fundamentals/