Improved LAG algorithm type

PeterUK
PeterUK Posts: 2,706  Guru Member
First Anniversary 10 Comments Friend Collector First Answer

So this is what everyone wants and no one had done it.

Put simple this algorithm would be like a deck of cards (packets) with players (ports in a LAG) and all you do is send one card down to one player next card to the next player and loop back around.

It will work great regardless of packet size

1 votes

Active · Last Updated

Comments

  • Zyxel_Melen
    Zyxel_Melen Posts: 1,629  Zyxel Employee
    First Anniversary 10 Comments Friend Collector First Answer

    Hi @PeterUK ,

    Thanks for your input. I would like to confirm:

    1. Do you mean the criteria? Like adding L3+L2 criteria or something like that?
    2. If not, do you mean to enhance the effect of the hashing algorithm?

    Zyxel Melen

  • PeterUK
    PeterUK Posts: 2,706  Guru Member
    First Anniversary 10 Comments Friend Collector First Answer
    edited January 18

    Hi Melen thanks for your interest

    As in it does not matter what the packet is or the size it just passes them down the the ports as they come in. So let say you did a two port LAG as packets come in to be sent out the LAG it alternates sending packets down the ports bit like a RAID 0.

    This would allow a single connection thread on a 2.5Gb port to fully use the LAG two 1Gb ports without being stuck to one port and the other end will just accept this.

  • Zyxel_Melen
    Zyxel_Melen Posts: 1,629  Zyxel Employee
    First Anniversary 10 Comments Friend Collector First Answer

    Hi @PeterUK,

    Thanks for the detailed information. It seems like you want a "load balance" algorithm. However, it is not ideally suited for network switches due to the behavior of chipset design. The suitable algorithm of chipset design for network switches is "load sharing."

    The "load balance" algorithm can lead to packet order issues in network environments, especially with UDP traffic. If packets are sent out of order, it can result in significant data stream corruption and traffic drops. To be aware, the roles of RAID and switch are different, one is storing data and one is forwarding data. While this might not be a critical issue for RAID 0, it poses a considerable challenge for switches, where maintaining packet order is crucial. That's why the switch chipset design uses "load sharing" instead of the "load balance" algorithm.

    Zyxel Melen

  • PeterUK
    PeterUK Posts: 2,706  Guru Member
    First Anniversary 10 Comments Friend Collector First Answer
    edited January 26

    I don't really see an issue with packet order as Docsis dose a single thread faster then one channel using many channels.

    I guess you could do a mix "load sharing" UDP and other protocols "load balance" for TCP algorithm

    or "load sharing" with TCP/UDP ports with IP source/destination this would still limit single thread but multi thread using another source port to same destination port will improve


  • Zyxel_Melen
    Zyxel_Melen Posts: 1,629  Zyxel Employee
    First Anniversary 10 Comments Friend Collector First Answer
    edited February 16

    Hi @PeterUK,

    Thanks for keeping sharing your ideas. Honestly, this idea is related to chipset support. However, there's a chance that the traffic passes on the same port even if using TCP/UDP ports with IP source/destination algorithm. We will keep monitoring this idea.

    Zyxel Melen