3

With my current design I have a DL380 Gen9 running ESXi. It has redundant NICs cabled to two different switches. Those switches in turn are both cabled to a third switch that allows my workstation to be connected. My testing so far has shown the following.

Redundant networking diagram

  • If cable C is unplugged, host is still available.
  • If cable D is unplugged, host is still available.
  • if either one of the M4100-50Gs is powered off, host is still available.
  • if cable A is unplugged, host is no longer available.
  • cable B hasn't been tested.

Is there any way with the current hardware that failing cable A could still leave the DL380 accessible to the workstation?

MadHatter
  • 79,770
  • 20
  • 184
  • 232
Digital ink
  • 500
  • 1
  • 10
  • 23
  • First of all you should connect the two core switches (NetGear M4100...) than unplugging cable A or B won't matter. Don't know if it's viable in your scenario but you should have 2 nics on workstation and redound the access layer (switch NetGear GS748Tv5) so the client has 2 path to the server. if isn't possible to redound the access layer your setup seems decent to me. – eldblz Aug 31 '15 at 20:52
  • @eldblz: Firstly, you don't provide redundant servers and services by using redundant network cards in your client machines. Secondly, what does `redound` mean? – joeqwerty Aug 31 '15 at 20:54
  • @joeqwerty why not? two nics on the same network segment on different switchs are perfectly fine if you use network teaming: redound is the verb for redundancy no? Should mean "Duplicate" but for redundancy purposes. – eldblz Aug 31 '15 at 20:57
  • So scale that out to 50 or 500 clients that need to access services provided by the virtual machines on that host. That's not how you do redundancy and high availability. Show me a single example of an enterprise using multiple network cards in their client machines to provide highly available and redundant access to services. – joeqwerty Aug 31 '15 at 20:59
  • 1
    If all switches being used support STP You should be able to lose either cable A or B and still function. There just might be a temporary loss in traffic being passed while STP is being recalculated (depending on how you have it configured). – Gene Aug 31 '15 at 21:03
  • @joeqwerty i never said (nor the question asked) it's a scalable solution but if that particular workstation is critical to production can be done. I can show you many example when a workstation (must be always connected to a server cluster in order to get blueprints) and the same workstation hold control for industrial laser cutter if the workstation lose connectivity to the server the machinery stops (don't know where cut) the production will stop (company losing money). And no unfortunately those expensive machinery (laser cutter, press etc) does not support ip connection (for now). – eldblz Aug 31 '15 at 21:04
  • It could be that I was looking at the bigger picture. If this is a one-off and the end goal is to only provide redundant and highly available access for this one client, then having multiple network cards in said client would probably do the trick. – joeqwerty Aug 31 '15 at 21:07
  • @eldblz I didn't do a NIC team on the workstation specifically because I want to simulate the environment of a larger node network. Like joeqwerty mentioned, I don't think you would ever see client machines with NiC teams in larger networks. Also, I didn't cable the two core switches together because I didn't feel I knew enough about the protocols that make looped connections work properly. Spanning tree? – Digital ink Aug 31 '15 at 21:07
  • Donovan ok sorry for my comment than. However you should consider connecting your two core switches together and enabling spanning tree. as @Gene mentioned. STP is defenetly a best practice in an enterprise enviroment and it's a scalable solution. Look for hierarchical network design: http://www.omnisecu.com/cisco-certified-network-associate-ccna/three-tier-hierarchical-network-model.php – eldblz Aug 31 '15 at 21:11
  • @eldblz thanks for the link. I don't have the gear to simulate that setup but I get the concept. – Digital ink Aug 31 '15 at 21:46

1 Answers1

4

Assuming there isn't a firewall or router between the end users and the server you could employ STP (spanning tree protocol) and the following layout:

proposed network layout

This would give you the redundancy you are seeking. If a single "core switch" (the m4100-50) goes offline traffic should be able to keep flowing.

Depending on how STP is configured (what type of STP and specific configuration options) and how many switches are involved there is some time required for STP to recalculate, but at this point you're looking at recovery times less than a minute, possibly even just a few seconds.

Additionally, note the inclusion of a link between the "core switches" (the m4100s). If these switches are connected you could lose the left links (2) between the core and edge switches and the B link between the VM host and core switches and still have connectivity.

Gene
  • 3,663
  • 20
  • 39
  • thank you. It looks like STP is the missing link. I think next I'll read up on Netgear's implementation. – Digital ink Aug 31 '15 at 21:47
  • It was a few years ago, but at the time Netgear's implementations of all spanning tree variants was buggy as hell. We ended up paying a bit more and bought Dell PowerConnects in our access layer instead. HPs are also in the "quality but not ludicrously priced" category. Basically, anything but Cisco is generally fine for the access layer. – rmalayter Aug 31 '15 at 23:23
  • @rmalayter, were you running all Netgear hardware, or were you mixing and matching? I'd imagine that STP using only Netgear switches should work fine. – Gene Sep 01 '15 at 13:46
  • @Donovan, rmalayter brings up a good point. The usefulness and stability of STP may vary between vendors, and for "prosumer" hardware it might not be all that great. Make sure you do plenty of testing. If you end up looking into different hardware I also recommend looking at HP ProCurve equipment. They make good stuff. – Gene Sep 01 '15 at 13:47
  • @Gene, we were not using all Netgear equipment. But RSTP is a standardized protocol (IEEE 802.1w). Getting STP correct is table stakes for making switches of any type. You have to handle the case where someone plugs in another switch to a port sensibly, no matter what. Even Cisco can't get away with "must use all Cisco in your network", and NetGear is no Cisco. – rmalayter Sep 02 '15 at 03:20
  • Oh, I'm not saying that it's okay for any vendor to require that only their gear be used, just saying that netgear products probably play better with their own gear. It's not okay if it does, that just might be how it is. – Gene Sep 02 '15 at 03:43
  • 1
    For the level of operation we have the Netgear switches have done fine. I've never asked much of them. Just simple switching and light VLANing. I guess I could say the worst thing about the network is its all Netgear but then the best thing about it is that... Its all Netgear. Heh. I don't have any formal networking training so I suppose there will be lots of testing in my future. Thanks everyone. – Digital ink Sep 02 '15 at 16:59
  • +1 for "there will be lots of testing in my future" – Gene Sep 02 '15 at 17:00