Yes, you are correct, the command no switchport is not used in this configuration. This means that this EtherChannel setup is a Layer 2 Etherchannel connection. That is why DTP is running, and that is why you are able to configure the link as a trunk just by configuring a trunk port on only one end of the EtherChannel as stated in the previous post. All of these features are Layer 2 features.
If you were to use the no switchport command, you would be configuring a Layer 3 EtherChannel link. For more information about that, take a look at this lesson:
It looks the load is zero for both ports. I want to know the amount of load so I can choose whatās best method of load balance to have (i.e. src-mac, dest-mac, ā¦etc.). But as you can see the results came back zero. Do you know a better way to see how much traffic on this port?
Also, this switch has several hosts connected to it, and itās connected to another switch via fiber etherchannel, it also the main source of internet.
Another question I have is confirm my understanding that itās better to select a protocol such as LACP or PAgP if you want to perform etherchanel between a switch and server rather than On mode. And if weed to etherchannel two switches then no problem to choose the mode ON. Please confirm my understanding. Thanks for your help!
Hmm, thatās strange. I labbed this up to see what kind of traffic I could see on my switches, but I too see that both the load and the number of bits remains zero. Iām not sure why. Iāll continue to look into it howeverā¦
To answer your question, the load balancing method you should use depends upon your topology and your expected traffic coming in on each end of the portchannel. Take a look at this post to see how you should approach this issue:
Now once you implement your load balancing algorithm, how do you check to see if traffic is sufficiently balanced? On older platforms, you could use the test etherchannel load-balance interface port-channel command with appropriate IP and MAC address parameters, and this would tell you out of which port such a packet/frame would be forwarded. But on newer devices this is unavailable. The best way to determine this, and to gain full visibility is to use a network management system with SNMP such as Observium, that will show you over time the actual traffic on each individual physical link. No CLI command comes close to such clarity. However, for a quick and dirty method, take a look at the interface statistics of each physical interface. There you can determine if the traffic on each interface is comparable, or if one link is being favored over another.
I am looking for some help regarding the LACP bundling time, and how I can speed that up.
This question has cropped up a few times for me, but I can find little help online regarding tweaking LACP bundle times.
I have seen that some devices support LACP Fast Rate - however this seems to be more of a keep alive once the bundle has been established.
Some people are suggesting actve/active is fast than active/standby - but I cannot find any support for this.
In my own research I can only find the LACP-Max-Bundle combined with the LACP fast-switchover command to be the only effective way of speeding this up .
Does anyone know of anything specific I am missing that can speed up the link-budling process?
First of all, fast rate does indeed speed things up (if it is supported) since LACP PDUs are sent every second rather than every 30 seconds, which ensures that you have a comparatively quick negotiation time. Even though it is the interval of the keepalive, it also ensures initial negotiation is much less than 30 seconds.
The active/active suggestion makes sense since if you have active/passive, the active side waits several seconds to āhearā a response from the other sideā¦ However, I have labbed this up, and at least in CML, I have found that there is little difference in the bundling time.
The max-bundle and fast switchover commands are used somewhat differently and have to do with creating active and hot-standby links within the same bundle. You can find out more about those at the NEtworkLessons notes about Etherchanel - max-bundle and [Etherchannel - fast-switchover]https://notes.networklessons.com/etherchannel-fast-switchover). However, if you have experimented with these and found that it speeds up initial LACP activation, then Iād be interested in further hearing your findings.
It seems that beyond what you have already looked at, as far as I know at least, there is no further method of speeding up LACP negotiation. If anyone finds something that works, feel free to share it!
This may seem like a stupid question but does the etherchannel load balance have to match on both sides of the link ? Or can I have different load balance methods configured on each side of the link ? For example if I have switch 1 on one side and switch 2 on the other side can I choose a different load balance method on each side ?
Thatās not a stupid question at all but it is important to understand how load balancing in EtherChannel works.
The load balancing configured on a switch affects only outbound traffic. So the algorithm chosen on one end of the EtherChannel has no effect on the one chosen on the other end of the EtherChannel. So you can have any method configured on each end.
However, you should keep in mind that some applications like VoIP may suffer if you have different paths for their packets in opposite directions. Although itās not mandatory, it may be a good idea to configure load balancing algorithms such that they have return traffic take the same path.
Can we purchase two P2P circuits ( simple L2 ) from two different vendors ( ATT and Verizon for example ) and Bundle them together terminating on our equipment between two of our datacenters and give that LAG interfaces a /30 to run OSPF over it ?
Is that a possibility and can you guys think of any concerns ?
NOTE : Just sticking with one vendor is giving us intermittent connection drops between our Datacenters and some applications donāt like frequent connection drops.
When you create a Layer 2 link aggregation, it is necessary to configure the aggregation parameters on both ends of the link. For example, when you create a bundle of any type (Etherchannel, LACP, PAgP, manual config, etc) you must configure the interfaces on both switches on which those bundles terminate. Otherwise the LAG will not form.
So the question here is, where do your P2P links terminate and what kind of service are you purchasing from your ISPs? If you are given a clean L2 link (MetroEthernet) from each ISP and you have your own switches on both ends of those links, then yes you can do it. The prerequisite is that the switches on each end see each link as an L2 link to your switch at the remote site. For example, take a look at this diagram:
SW1 and SW2 see the connections on their Gi0/1 and Gi0/2 interfaces as if they were directly connected using two patch cables. If that is the case from the point of view of those switches, then the answer is yes, you can bundle those two links together without any specialized configurations.
Just keep in mind that based on the traffic patterns you can expect between the two datacenters, you should consider the best load balancing algorithms to use to ensure the most efficient use of the links.
There are 3 ways of configuring port-channels as youāve advised, PAgP, LACP and Manual.
What does the following command channel-group X (X = a number) do on an interface then? This also builds a port-channel but how is this different to the above 3?
The command channel-group X in interface configuration mode is incomplete. Notice the following:
SW1(config)#inter gig 0/1
SW1(config-if)#channel-group 1 ?
mode Etherchannel Mode of the interface
SW1(config-if)#channel-group 1
% Incomplete command.
SW1(config-if)#
The keyword mode must follow along with the commands that will specify whether you are using PAgP, LACP, and manual. So you must explicitly configure one of the possible methods given.
What would be the effect if we had switch interfaces configured in a port-channel (with 2 interfaces) but on the other side e.g. VMWare ESX host or even another switch had 2 individual interfaces and not in a port channel?
Regardless of what device is on the other end, if we have EtherChannel configured correctly on one end, and not on the other, then you will have some unpredictable behavior, and this depends upon your configuration.
Remember that Ethernchannel can be either manually set or negotiated using PAgP or LACP. In each case you may have a little bit of a different behavior.
In the case of a manual configuration, if you set it up on the switch, no matter what you do on the other end, it will look like everything is OK. It has no error messages because thereās no way to detect whatās going on at the other end. HOwever, if you have connected to a switch, the other switch will treat the links as simply two links from the same switch, so using STP, it will block one of the links. So you would have a working link, but youād be using only one of the two physical connections at any one time.
If youāre using PAgP or LACP, then there will be a negotiation problem and an error will appear on your switch. Even so, you still may have connectivity since the physical interfaces stay up. But only via each individual physical interface and not via the port-channel.
So you will have connectivity, but it may be unpredictable as to which link transmission will occur.
When configuring anything on an EtherChannel bundle, we always apply that to the PortChannel interface, and not on the physical interfaces. The only time we need to apply a parameter on the physical interfaces is when we are configuring UDLD. Take a look at this NetworkLessons note on applying configurations to an EtherChannel interface for more details.
As far as I can tell āport-channel load-balanceā is a global configuration command, thus it affects all active EtherChannels.
SW1 is configured to load balance based on the source whereas SW3 is configured to load balance based on the destination.
Now letās assume the PCs want to interact with the server on L2 or L3.
Two scenarios:
SW2 is configured to load balance on the source MAC or IP
SW2 is configured to load balance on the destination MAC or IP
In scenario 1, both physical links from SW2 towards SW3 will be used. However only one physical link will be used from SW2 to SW1 as the source is permanently the same.
In scenario 2, only one physical link from SW2 towards SW3 will be used for the reversed reason (constantly same destination). On the other hand from SW2 to SW1 both physical links will be used.
Am I right saying that? Which load-balance algorithm would be preferred to use on SW2? I assume taking source and destination MAC or IP into the algorithm would solve this?
I understand the dilemma. For SW2, both scenarios will cause a less-than-optimal load-balancing scenario. By default, the source MAC address is used, so even if you donāt configure any load balancing, it will use the source MAC to load balance. One thing to note here is that load balancing configured on a particular switch for port channels always affects the outgoing traffic.
The solution to this particular scenario would be to base load balancing on both the source and the destination MAC or IP address. That way SW2 will more evenly distribute traffic across the port channel.
