Slightly OT: Network switch failover?



  • Just curious about this right now.  I have a "core" switch I'm going to be working with, and I have a slightly lesser model I'll be keeping around in case of a failure.  I'm wondering, is anyone aware of a method to make a failover more or less seemless, other than me getting a page that the "network is down", drive the half hour down to the data center, spend more time figuring out that the switch has failed, and manually unplug all of the cables from the core switch and into the backup, and wait as the arp tables get updated?



  • About how many ports/servers are we talking? You could double the links of the servers (team them as failoverteam) and hook up each link to one of the switches. Add 2 pfSenses in failoverconfig with each LAN nic connected to one of the switches. Crossconnect the switches to each other. Make sure both switches support at least spanning tree protocol. With this kind of setup you should be able to power down any of the switches (If I didn't miss something  ;) ).



  • Sorry Just butting in…

    That is how I currently work. I have what is called a "Core Stack" with all switches attached to it with a STP path cost of 4. Then I have a backup Stack with all switches attached to it with a STP path cost of 100 and an uplink between both stacks.

    This allows me to shut down any switch (including any of the core switches) with a failover of under 2 seconds (1 ping) when using Rapid Spanning Tree.

    Only server I don't have yet is pfsense, didnt know it was possible to create a failover team, gotta look into that!

    Hope this helps!



  • Right now about 30.  I just bought an 80 port switch to give me room to grow, and was tempting to get 1-2 lesser quality switches to failover onto in case the core fails.

    "Double the links of the servers."  Do you mean use two nics in each server and plug into both switches and then turn around and use STP on the switches?  ???  Mac address/IP address conflicts abound here.  STP would kill off the packet storm that would normally ensue, but I don't see how you're wiring this. :(

    To complicate matters, I'm using 2 switches as is without the failover.  My front-facing 10/100 switch that is highly managed (where machines get public IP addresses), and a lesser managed gigabit backend switch where MySQL transactions, filesystem exports, backups, etc. tack place (private IP addresses), and each system has 2 nics already.

    I'd esstentially be going from 2 switches to 4, each in a failover pair, but I'm having trouble picturing the wiring here.  Two switches support stp.  Use a cross-cable or an uplink port to hook them together in two spots, enable stp.  I plug a system into switch 1…now how do I hook it to switch two without bringing another device into the mix?



  • Server
      |  |________________
      |                  |
    switch1–----------- switch2
      |                  |
    pfs1-----(sync)------pfs2
      |___________________|
                  |
                  WAN

    I am using some HP servers at a customer with intel nics. The driver has a utility to team adapters. You can set them to different modes like failover, loadbalancing and so on. In your case I would set them to failover. This way only one of the nics will hold the traffic. The traffic will switch to the other nic if one of the links dies. IP will remain the same on both nics.
    Check out this link for details: http://www.intel.com/support/network/sb/cs-009747.htm



  • Ah, so in order to do this, I'd have to have 4 nics in each server, 2 frontside, 2 backside.  Plus, those nics would have to support team failover on whatever OS they're using (mostly MacOS X Servers, and now I'm adding in about 10-11 FreeBSD servers).

    Yay…that's a lot of nics.  Some of these boxes don't even have that many pci slots.  I know for fact that several of the on-board gigabit ports are supposed to replace the need for pci slots, so they only have 1-2 slots available.

    Switch failover may be a lost cause in my environment.  I may just have to be ready with the swappable switch sitting there as a just-in-case measure. :(


Locked