Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

We have an office network that consists of the following:

1x Vigor 2950 5-port (WAN Load Balancer) 2x Netgear 24-port Managed Switch FSM726 1x PowerConnect 2724 24-port

  • The Vigor has our two ADSL lines hanging off it.
  • Our patch panel connects into the 2 Netgear (for all desktops, laptops, etc...)
  • All servers plug in to the PowerConnect

Currently the configuration is:

  • Netgear 1 connected to Netgear 2 using GB ports
  • PowerConnect connected to Netgear 2 using GB ports
  • Netgear 2 connected to Vigor 2950 using GB ports

Basically, the question I have is this the correct way we should be doing it? We had an instance last week when a user was copying 10GBs of large files from a server on the PowerConnect to his machine on Netgear 1 and it basically killed the network for everyone else except him.

Should I infact be connecting each Netgear and the PowerConnect into the Vigor instead?

I'm not sure what the rules are for connecting multiple switches together and I don't seem to be able to find anything good on Google.

Thanks.

Niklas

share|improve this question
add comment

4 Answers

If you've got more than one connection between the same two switches, either:

a) Make sure that you're using spanning tree, or
b) Don't do it.

Remember, the LAN ports on the Vigor box count as a switch as well. I'm not sure what sort of bandwidth management options the NetGear switches give you, but you may also want to look into that. As far as the actual setup goes, it looks pretty solid.

share|improve this answer
    
+1 for STP. Recently had an organisation wide black-out when someone plugged a cable into the wrong port. –  sybreon Sep 9 '09 at 11:00
    
I've done something similar myself. Once the lack of STP bites you, you tend to stay bit... –  RainyRat Sep 9 '09 at 11:10
    
So basically daisy chaining the switches are the correct way to do it rather than connecting the Netgears and PowerConnect into 3 ports on the Vigor? Especially as the Netgears are the only ones that have spanning tree support? Thanks for your feedback. –  Niklas Sep 9 '09 at 13:27
    
STP doesn't work that great on my PowerConnects, or between different brands. See my answer about daisy chaining. –  kmarsh Sep 9 '09 at 13:39
add comment

The proper way of doing it is:

  1. Don't use Netgear and Dell PowerConnect switch for mission critical network operations (Been there, done that, moved on.) Their advanced features just don't work that well, especially when using more then one feature at once.

  2. Don't use a conglomeration of cheap switches for your network backbone. Invest in at least one large managed Layer 3 switch, with real phone tech support and 4 hour replacement. They exist and cost more for a very good reason.

  3. Don't use cheap switches for port aggregation to combine 10/100 and Gigabit Ethernet clients. They will drag down the performance of all Gigabit connections the moment the first 10 or 100Mbit is connected.

  4. Now that you have real equipment that doesn't choke, use EtherChannel (siamesed ports) or Stacking to connect backbone switches together. This will allow more than one user full Gigabit throughput internally.

  5. As RainyRat said, implement Spanning tree on EVERYTHING, even if it slows down recognition of new devices (30 sec instead of 3 sec).

As you have already discovered, cheap SOHO switches simply don't have the internal backbone to handle serious network traffic. Daisy-chaining them multiplies their limitations.

EDIT: If you can't afford that, you can try: EtherChannel the two NetGear switches together with a 2xGigabit link, and turn STP on. You can use 10/100 ports to limit your power user's throughput.

The PowerConnect is a managed switch, but I have found difficulties in utilizing more than one managed feature at a time. You can try to STP on the PowerConnect and EtherChannel to the NetGears, but I'm not optimistic about throughput. When I tried fixed port speed plus VLANs on my PowerConnects, they bricked and had to be hard reset.

share|improve this answer
    
Upper management couldn't understand or believe the price of good managed switches + tech support. Explaining 4 hour replacement helped, but they still didn't get it. Explaining "I can call tech support, put in a research request how to connect X # of switches with Y # of 4-port EtherChannels, STP+ loop-back protection and lock out physical cross-connects of VLANs, and get the correct answer back in a couple of days", that got through. –  kmarsh Sep 9 '09 at 13:46
    
Hi kmarsh, Thanks for the feedback. I appreciate that in an ideal world we should all be using top of the range kit, but if in the short term we can't what would be the best (given the circumstances) way to do it? Thanks in advance. –  Niklas Sep 9 '09 at 14:30
    
I don't think this answer is actually helpful - whilst we would all like to be able to replace kit, most people can't to solve a simple problem –  Neil Middleton Sep 9 '09 at 14:33
2  
I never said "top of the range kit". HP ProCurve costs 1/10 of Cisco equipment and gives you what you need. See edit above for other ideas. –  kmarsh Sep 9 '09 at 14:53
    
Hi kmarsh, thanks for the edit. So if I EtherChannel the two Netgear's together I then link one of the NetGears to the Vigor and then PowerConnect to the Vigor? Thanks. –  Niklas Sep 9 '09 at 15:25
show 1 more comment

If the user is on one switch and the server is on another switch, it doesn't really matter how the switches are inter-connected. If the problem is going to manifest itself it's going to do so whether or not the switches are plugged into each other or plugged into the LAN ports on the router.

What you need to do as a first step is look at the port configuration (speed and duplex) between the switches, ensure that you have a loop free topology, and check the inter-switch links for saturation.

A switch loop isn't likely as it wouldn't manifest itself only during the copy operation. It would exist all of the time and it would make the network virtually unusable at any time and all of the time. A switch loop isn't something that corrects itself. A switch loop can only be corrected by removing the loop or implementing the Spanning Tree protocol.

share|improve this answer
add comment

This is is an easy solution that you have over complicated and people telling you that you need to go out and spend more money are not being very helpful.

First you have to make everyone aware of how much, or little bandwidth is available. This is sometimes a good way to figure out who the bandwidth hogs are in the office. You need to sometimes notify people what they are doing wrong so they at least know it's not good behavior. Transferring 10GB of files during business hours while the rest of the office needs the network is bad, mkay?

But just teaching good beahvior is only the start. Take a couple ports on any of your switches and make it so the data hogs live on those ports on a single switch and that they are not saturating your made-up backplane links. If you can keep both machines of the data hog connected to a single switch and the rest of the non-data hogs on the other switch, your problem is solved. You could link together more expensive switches with real interconnect cables and have up to a 32Gbps backplane but for the few times that data hog will do his deeds, I doubt it will justify the cost. Keep doing what you are doing, but take my advice and move the data hog to a single switch and don't let the large data transfers happen during business hours or across switches. Your network design is flawed is all, but it can be fixed very easy.

share|improve this answer
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.