to the question of choosing a routing protocol in an enterprise network.

[Tagir Temirgaliyev]

Let's consider a typical star topology, there is a head office R1 and
branches R2 R3 R4 are connected to it via one WAN link. in this case OSPF
fully provides the required functionality.

 

but now many companies already require increased fault tolerance and
connect 2 or more WAN links to each branch as in the figure. how does
OSPF work in this case? by default, it loads both channels equally if
they are equal. or if the channels are different then OSPF will use one
channel with a higher bandwidth and not load the second channel, the
second channel will be used only if the first one fails.

 

some companies stop there. or they buy two identical channels from two
different providers or use one but pay for both.
the option to buy, for example, 10 Mbit MPLS and 50 Mbit Internet
disappears.

solution can be PBR, SD-WAN, EIGRP and BGP
PBR can be an output, but the configuration becomes more complicated in
the case of multiple interfaces on the LAN as shown.
PBR will have to be applyed all LAN interfaces

 

about BGP, people usually immediately say a learned phrase from a
textbook - it's EGP and has poor convergence.
but look again at the picture. for example two links between R1 and R2
in this case, poor BGP convergence does not matter.
if one link fails, then both routers R1 and R2 instantly redirect traffic
to the remaining link.
yes, there is poor convergence, and compared to OSPF and the R4 router
will receive the update with some delay, but this will not play a
fundamental role since there is a second working link there.

[balaji.bandi]

There are PROS and Cons on each IGP protocol, where deploying very important. OSPF only doe equal-cost LB, EIGRP does unequal, BGP does in different mechanism with full control.

it all depends on the environment and design you choose. some time we deploy the technology things change, so we need to move to different IGP if possible or apply the tweaks to meet the goal of network requirements.

if this is full cisco EIGRP does your work, if Open standard, OSPF with PBR can do your LB/failover setup.

if you go further SD-WAN or BGP, you have whole control whatever way you can play your Traffic engineering, but cost model and device support vary.

as the technology trends become widely available now, so we can take advantage of that and meet the requirements demand today era.,

[Tagir Temirgaliyev]

there is a business task, to create an effective network infrastructure. not overpaying. so you have to either choose OSPF + PBR or BGP

SD-WAN allows traffic of various applications to pass through different channels, for example, Voice and Scada (ASUTP) through 10 Mbit MPLS and email file sharing Windows updates via the Internet 50 Mbit vendor.

EIGRP also allows you to direct traffic from different prefixes to different channels using the Offset list, but also binding to the vendor.

BGP + several ansible scripts for fixed configurations of traffic distribution over several channels looks like the best way out

[Joseph W. Doherty]

When you have multiple links, of different bandwidth capacities, much also depends on the disparity of the bandwidths.

With "normal" routing protocols, they don't take into account actual link loading.  So, for example, if you had one link of 40 Mbps and the other 50 Mbps, more often than not, you would do just about fine costing the two links the same to have ECMP.

If there was a large disparity, like your example of 10 Mbps vs. 50 Mbps, also more often than not, again without dynamic load balancing, you normally do better just having the higher bandwidth path used leaving the (much) lower bandwidth path for backup.  Why?  Well if you direct some flows to the (much) lower bandwidth link, they will also be limited to that link's bandwidth.  Consider that applications might well "see" a signicant transfer rate difference depending on which link the flow ended up on.

In the past, I've use OER/PfR to dynamically load balance.  Using it, you can best leverage something like your 10 & 50 Mbps links, because a high rate flow that started on the 10 Mbps link may be shifted to the 50 Mbps link.

Another advantage to something like OER/PfR, if the hub site has two multi-point connections to spokes, traffic to the hub can congest at the cloud connection to hub.  This is something else OER/PfR can detect, and shift traffic around to obtain the best overall network performance.

(BTW, SD-WAN might be the current "gen" of OER/PfR, but I've not used it.)