We help IT Professionals succeed at work.

I want to aggregate a link between 2 switches - but I am unsure if it will work?

Medium Priority
761 Views
Last Modified: 2012-06-27
So I have 2 Cisco SLM2048 switches, and I would like for them to use link aggregation to talk to eachother to get a 2GBit link between them.

Is it as simple as creating 1 group of aggregated ports on each switch and just connect the cables?

I have aggregated before, but that is always to 2 network cards, I have never blindly just connected 2 groups of links and know if it works.

Will it or won't it?
Comment
Watch Question

Commented:
it will ;-))
Commented:
Gugro's right, it will work (but you can't expect 250 points per word for an answer :-D).  Just make sure that the ports are configured the same (trunk mode, vlans allowed or configured, speed and duplex settings) and it should come up.  There is really nothing more to it.

Author

Commented:
Ok, so we tried, and we had a success. Thanks for backing my theory up.

However (while i have some trunking people on the line, please bare with me, you will both get the points).


I have 4 network ports on my server:
2 on the mother board,
and a PCI card that has 2 NIC's on them.

I currently use the PCI running as a static address.
This sits on my domain controller, acting DHCP and DNS.

I used the included software from the broadcom suite to aggregate/link the 2 mother board cards to a 2GB connection. This was a pretty simple setup, basically click on through and select which cards you want.
The only option that made me think was WHAT type of link I wanted.
I could go with either "Smart load balancing and failover" or "802.3ad LACP".
I figured (perhaps naively) that the pre-selected smart option should work the best (I'm plugging this in to a Cisco SLM 2048 managed gigabit switch).

So long story short; I created a virtual network card with the linking/aggregation, assigned it the AD's IP, disabled the normal NIC, and restarted everything.

It worked fine (it seemed).
All went well, transfer speeds were great (60+ MB/sec), internet got DNS requests just fine, etc.

Then suddenly it stopped.
About after an hour people started complaining that they could not get on to the internet and not reach the shared drives. My computer (of course) worked just fine, as did my assistants and most other people.
But the ones who had the problem could simply not get a DHCP address.
I screwed around with it a bit, and it turned out they DID get DHCP addresses, but the NIC's acted as if they didn't have an address.

i could see that they registered in DHCP, and that they had an IP, but I could not ping ANYTHING, nothing worked, it was like they did not have an IP address at all even though they did.

I restarted everything like 3 times, routers, switches, servers, etc.
Nothing, still super weird.

Eventually I decided to kill off the teaming, and enabled the old NIC.
All was kind of back to normal.
Half of the office (of about 60 people) could work as normal, but other half couldn't.
It was not related to sub net (I only use 1), it was not that one switch worked and others didn't, they were connected to switches that had working people on it. It was random people not getting connections working with no common denominator.

To solve it I had to kill the aggregation setting for the two ports that had previously had the linked nic's in them, activate the old original NIC, and then it worked again.

This makes me wonder if it was the switch all along, but then why did it work for a while and not later?
There are only so many settings that the aggregation settings has to tinker with on the switch, so I don;t feel like I did some messup in the settings.
What difference should 2 empty ports in a switch do, even though they are linked/aggregated, they were not connected?

Long story off topic, I know, but this is really important that I figure out for my sanity...

Commented:
I cannot tell why a teaming set-up on your server could affect traffic forwarding on your switch/router...  Unless the devices with the problems really didn't have IP's (they couldn't reach the DHCP server), but it seamed that they did (like Automatic Private IP Addressing (APIPA) implemented by Windows when no DHCP server reply.  This could be caused by a malfunctioning port-channel:  Requests received/transmitted over 1 link in the channel reached it's destination, while traffic on the other didn't.  And since port-channel can loadshare based on src/dest MAC, it could be the same machines affected, while others worked, whether you rebooted or not.

When configuring a port-channel to a switch, try the LACP option:  It is a standard I know Cisco supports, and you should have less hassle than the (proprietary, I assume)"Smart Load-balancing and Failover" the NIC-manufacturer provides.  But I cannot guarantee that this was the cause of your problems, although you did get it going again when you changed the port-channel settings...

Also note that you have to configure port-channel on the switch's ports as well...

Hope this helps!

Author

Commented:
I don't understand 'configure port channel'. What do you mean by that?

Well, I was thinking that since this is a pretty standard server (Dell power edge T610), and the NIC's are pretty standard too (broadcom server class), that somehow the 'standard' choice of protocol would be compatible with Cisco and just work.

There are so few settings that you can screw around with (unless you want to go in and manually screw around) that I thought they'd have their stuff together by now.

What was strange to me was that some people worked just fine, and others didn't work at all. I just couldn't figure out what the problem could be. usually it works or it doesn't.

Commented:
I apologise, the terminologies differ sometimes:  A port-channel is the virtual link that forms if you aggregate more than one link.  Some vendors (I think it's HP or Dell) even call it trunking (not to be confused with 802.1q trunks).  It is also called an EtherChannel (when it consists of Ethernet links - n x 10Mbps), Fast EtherChannel (n x 100Mbps) or Gigabit EtherChannel (n x 1000Mbps).

Cisco, by default, have any port-channelling disabled, so you need to configure a port-channel if you want one.  Else the switch will treat the ports as normal.

Each vendor have their own preference, and these are not always compatible with each other.  So you need to know what the NICs are capable of, what the server and OS will allow, and what the network can provide, before you can determine what solution will work best.  I assume that's why I still have a job... ;-)

Explore More ContentExplore courses, solutions, and other research materials related to this topic.