ful lmesh

[bluesea2010]

Hi,

Ospf running , what is the pros and cons for the below full mesh topology ? . ospf type should be broadcast or point to point ? 

Thanks

[MHM Cisco World]

P2p, two l3 sw connect via direct link then p2p even if link is ethernet still only two peer is connected to this broadcast link.

[paul driver]

Hello
From that diagram IF you have auto-cost reference BW to accommodate those 25/40 GB interfaces then your routing by default won’t be optimal as traffic between a switch on bottom left to top right will route via the 40gb links instead going direct over the 25gb connections, unless you manually append the interface costing

[Joseph W. Doherty]

We're lacking much information, concerning OSPF's setup to answer your questions.  For example, how those links are OSPF costed, whether all these links are within the same OSPF area, etc.

Although you show a full mesh, but as you also show "inter" links have less bandwidth then the "outer" links, from OSPF's perspective, you might have a ring topology as your 40G links being best path, and the 25G links being "backups".

Assuming links are Ethernet, but two ends are logically p2p, such as using a /30 or /31 network, then OSPF should be configured to treat the link as p2p.

[bluesea2010]

Hi,

The ospf cost

We're lacking much information, concerning OSPF's setup to answer your questions. For example, how those links are OSPF costed,

Whether all these links are within the same OSPF area , yes all are in same area

cost is auto , if I set cost same for inner and outer links , what could be the problem 

Thanks 

[Joseph W. Doherty]

Cisco's default for auto OSPF costing is the "fastest" interface is 100Mbps, so all your links 100Mbps, or faster, will be costed as "1".

[MHM Cisco World]

from cisco
""In Data Center environments, there are typically Gigabit Ethernet links and even 10 Gigabit Ethernet links. To address this problem, you can change the reference bandwidth to 10 Gbps with the auto-cost reference-bandwidth 10000 command. You should use this change consistently in the OSPF network.""

[Joseph W. Doherty]

". . . you can change the reference bandwidth to 10 Gbps with the auto-cost reference-bandwidth 10000 command."

Yea, but, the OP shows 25 and 40 Gbps links, i.e. a 10 Gbps OSPF would treat 10 Gbps interfaces, or faster, all the same.

[bluesea2010]

Hi @Joseph W. Doherty 

In that case   these type of ring topology is recommended ? 

is there any issue of leaving auto reference 

Thanks

[Joseph W. Doherty]

"In that case these type of ring topology is recommended ? "

Unfortunately, that an "it depends" answers.

Using a full mesh, rather than the "ring" would reduce some inter switch latency.  However, your outer links have almost double the bandwidth of the inter links, so cannot say which way (i.e. ring or mesh) is likely to congest (if it does, at all).

"is there any issue of leaving auto reference "

Well, if using the default base cost, all your 25 and 40 G links will be equal, i.e. you should effectively have a full mesh.  Since the links are not the same bandwidth, without knowing traffic patterns, cannot say whether this is good or bad.

Also, though, you should consider the rest of your topology, i.e. do you want OSPF to distinguish between links with more than 100 Mbps?  (NB: many would likely increase the reference bandwidth value.)