Cisco Support Community
Showing results for 
Search instead for 
Did you mean: 

Welcome to Cisco Support Community. We would love to have your feedback.

For an introduction to the new site, click here. If you'd prefer to explore, try our test area to get started. And see here for current known issues.

New Member

cat 2950g-12 stacking limit


I wonder about cat 2950G-12 and stacking count limit ( no CMS, only data communication )

switch (GBIC LX) ---(Single mode Fiber)--- (GBIC LX) switch (GBIC LX)---(Single mode Fiber)--- (GBIC LX) switch (GBIC LX)..........about 20 switches serial.

*the distance between switches is 2 ~ 4Km.

*Subway envirnonment

is it possible ?

if possible then Layer 2 loop cabling(switch1 connect switch20) and STP(spanning tree protocol) can be used ? (20ea switches and total 80Km, Layer 2, STP Redundancy)


Re: cat 2950g-12 stacking limit

Do you have the hardware already, or know someone who does, so that you can test this in a lab setting first?

If you disable STP you could theoretically daisy-chain switches one after the other after the other as far as you want, as long as you do not create a loop. But eventually performance from the ends would become the issue, and this depends on your applications and how much data traffic your users produce.

You can create a loop from the switches on the ends, using a pair of ZX GBICs rather than LX/LH GBICs. If you create the loop, you must have STP enabled. But the default timers are designed around the assumptions of the Hello-Time interval = 2 seconds, and the Network Diameter = 7. You're talking about a network with diameter = 20!

You would need to make some changes to STP parameters on your STP root bridge/switch for STP to work in this scenario.

If you leave Hello-Time at 2 seconds, you can change the STP Max-Age from default of 20 to 46 seconds and the STP Forward-Time from default of 15 to 34 seconds. With these changes, expect traditional STP reconvergence in 68 to 114 seconds.

Even if you change the Hello-Time to 1 second (the minimum), you could only drop the corresponding Max-Age to 42 seconds and the STP Forward-Time to 32 seconds. This would result in traditional STP reconvergence in 64 to 106 seconds.

The key to getting STP reconvergence times down, besides using STP UplinkFast, is to reduce the Network Diameter size. If you can break up the 20 switches into smaller Layer 2 loops, then use some Layer 3 switches to connect the loops, you can lower the reconvergence times significantly. If you substitute a Catalyst 3750G-24TS switch for every 5th 2950G-12 in your original design, and buy five more pairs of ZX GBICs, you can create a pretty robust network. You would also need several extra pairs of fiber for those additional ZX connections.

Taking the Layer 3 idea to its logical extreme, if you were to build this network completely out of L3-capable switches (for example, Catalyst 3550-24-EMI), you would have no STP issues to contend with. And you could go back to that single pair of ZX GBICs to close the "loop". Run OSPF to have the network automatically adjust to network changes.

A very interesting design and cost evaluation problem you have posted here. I have a whole white board filled with ideas on this; and I will need a spreadsheet to total all the costs of the various approaches, and see which one gives the best performance for the most affordable price.

But it's been a long day, so I'll have to get to that tomorrow...


Re: cat 2950g-12 stacking limit

[My ISP decided they were going to reboot some of their routers last night, I lost connection to the Internet long enough that I could not add the following to my original post.]

In the meantime, if I had to guess, I would say the Layer 2 approach with the adjusted STP timers will be your best option. While the reconvergence time would be long relative to other designs, once you get the network up and running it shouldn't need to reconverge all that often. Spending lots of money to solve a problem that will only be a problem maybe 1% of the time is hard to justify, unless that 1% of downtime costs you more than the equipment that would have been used to build the better and faster solution.

Hope this helps.