Practical application of EtherChannel/link aggregation

geek4godgeek4god Member Posts: 187
Was wondering if there is a point of diminishing returns here. Trying to do some load balancing across three switches and in doing so I think I have created a bit of a bottleneck with the connection between them. Was just curious from a practical standpoint if there is a point of diminishing returns here.. Using gig pots here and I have plenty of ports I could use.. 10 ports would not be out of the question, just looking for some personal experience and suggestions..

Comments

  • vinbuckvinbuck Member Posts: 785
    Two questions,

    1) Are you trying to build one etherchannel across 3 switches?
    2) What type of switch, and what load balancing options do you have?

    Etherchannel's effectiveness all lies in selecting the right load balancing algorithm for the type of traffic you need to load balance. Sometimes using the default settings can lead to a lopsided distribution of traffic across the links.
    Cisco was my first networking love, but my "other" router is a Mikrotik...
  • geek4godgeek4god Member Posts: 187
    Sorry have to be careful with my terms, no actual load balancing.

    Had a Nortel stack die on me... I have a temp solution in place with some Adtran switches. Basically I have one core switch with my WAN connections, exchange box, and firewall plugged into it. Two other switches feed it. On those two switches my IDFs and the rest of my servers are distributed between the two. Saw some performances issues when I loaded down the core NetVanta 1544p so I tried to “balance” things a bit by off loading some of that to the other two..

    Really wanted to replace with Cisco 3750s in a stack, but I just don’t see that happening, because of the cost. So I will either use Cisco 3560s or a group of Adtran’s..

    Short term I wanted to increase the link aggregation that feeds the core switche, long term I am thinking about the permanent solution.. Looks like link aggregation tops out at 8 ports on 802.3ad so that answers part of my question. Was just curious if there was some diminishing returns before I started tying up ports..

    Part of this is the result of me having to come in because a loss of confidence in a vendor, so I am drinking from a fire hose on most of this. So forgive the noobness...
  • Forsaken_GAForsaken_GA Member Posts: 4,024
    Vin is exactly correct. No load balancing solution is going to give you perfect load balancing. What options you have all depends on your platform, and your distribution will be dependent on the type of traffic you're using. The more variance you can toss into the load balancing parameters, the better distribution you'll get. For example, if you're hashing based on source MAC, then the same MAC will take the same link every single time. If that source MAC is the one that's saturating your ports, then that's a bad choice to hash on. If most of your traffic is outbound to many ip's, then balancing based on destination IP is a good idea. Contrarily, if all your traffic is destined to one IP, using only destination IP is a bad choice.

    So you have to know what's available on the platform, and you have to know what kind of traffic is transiting the link. About the only time there's no point to etherchannel is if you have enough traffic between the same source ip and mac going to the same destination ip and mac that's saturating the link. Once that happens, your only real option is upgrading the circuit to a higher bandwidth one.

    The higher end platforms can usually load balance on layer 4 information as well, and given the random nature of TCP/UDP origin ports, will provide you the best load balancing distribution. That's pretty easy to find in a production environment, but very rare in a lab environment.
Sign In or Register to comment.