Performance of L3 Etherchannel Vs Separate load balanced L3 links

[Aaron_un]

Which one has higher performance/throughput/bandwidth for uplink to the next switch?

Layer 3 Etherchannel(2 links bundled) vs 2 separate equal cost load balanced(ospf) Layer 3 links?

I think the paraphrase to this question would be "OSPF equal cost load balancing algorithm or LACP"?

[balaji.bandi]

I think the paraphrase to this question would be "OSPF equal cost load balancing algorithm or LACP"?

It all depends on the requirement and design you looking to do and implement.  how you want to segment, I prefer to have fewer broadcast Laye 2 domains and STP Loops where possible.

Take example :

If your access switch connected to 2 Individual cores, you can not make it  Laye 2 Port-channel between disjoint Cores.

In this OSPF does the job for Equal Loadbalance.

[Aaron_un]

If your access switch connected to 2 Individual cores, you can not make it  Laye 2 Port-channel between disjoint Cores.

In this OSPF does the job for Equal Loadbalance.

---

no not layer 2, Layer 3!

let's make it simple, single switch to single switch

Which one has a higher throughput?

[Giuseppe Larosa · Accepted solution]

Hello @Aaron_un ,

etherchannel load balancing is used when you have a L3 port channel and CEF load balancing is used when you have ECMP.

CEF load balancing is flow based and it is an EXOR of less meaningful bits of Source IP address, Destination IP address and a seed value chosen at device startup.

Etherchannel load balancing is again flow based and if you use the IP source and IP destination in EXOR the results should be similar to those achieved by CEF load balancing.

In any case   a single IP flow is limited to the speed of a single member link with both solutions.

For some specific cases when using also the L4 ports in etherchannel load balancing you might get better more fair distribution of flows with bundle when compared to CEF ECMP.

Hope to help

Giuseppe

[balaji.bandi · Accepted solution]

The hashing operation can be performed on either MAC or IP addresses and can be based solely on the source or destination addresses, or both. Use the following command to configure frame distribution for all EtherChannel switch links.

more information can be found here 

https://www.cisco.com/c/en/us/support/docs/lan-switching/etherchannel/12023-4.html

[Georg Pauwen · Accepted solution]

Hello,

I would say the throughput is the same. The default load balancing algorithm for a layer 3 Etherchannel is src-dst-ip (the equivalent of per-destination load balancing on the router, which you would use for OSPF).

How this applies to your specific overall topology of course depends on the traffic flow in your network. But in general, there should not be much of a difference between both.

[Joseph W. Doherty]

"The default load balancing algorithm for a layer 3 Etherchannel is src-dst-ip . . ."

I recall (?) that (i.e. being the default) not being true for some/older platforms.  I would suggest, if hashing algorithm not already explicitly defined in config, to check what platform is actually using for its default.

[MHM Cisco World]

according to cisco the recovery is hug different between MEC and ECMP.

prefer MEC.

[paul driver]

Hello

TBH -I don think that shows the whole story - Looks like you've taken that from a cisco live HA architecture design doc - here

[Joseph W. Doherty]

As the other posters have noted, assuming your flows "randomize" across both links, Etherchannel and an IGP ECMP should about be equal.

As also noted in other posts, your have "options" for what to use for hashing attributes for Etherchannel.  This could provide better balancing than ECMP.

Not noted, though, depending on platform you may have a couple of options for an IGP ECMP to provide more/better utilization of your links.  First, some platforms support per-packet CEF load balancing.  This will often get you an almost 50/50 split, but as packets can be reordered, it's generally not recommended.  (BTW, this would also allow a flow to obtain more than one link's bandwidth.)

Second, some platforms support PfR which can dynamically load balance flows across your links, which Etherchannel nor ECMP normally do.

Another difference, between the two solutions, is how the IGP "sees" the metric for using the links, including when an individual link drops or joins the topology.  Etherchannel will been seen as just one link, but possibly with changes in its bandwidth if a link drops or joins the bundle.  I.e. the change might ripple though your L3 topology changing "best" end-to-end path.  With ECMP, end-to-end path cost shouldn't change.