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

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

Switch Design and sprawl

We have several schools using the 3524XL switches in a MDF and IDF scenario. We have 4-5 in the MDF and then 4-5 gig links out to the IDFs with 3-4 switches in ech IDF. We are looking to expand this with some new program that will have 5 jacks to each classroom and thus even more switch expansion.

I am wondering what type of limits do I need to be concerned with. Is there a limit to the number of switches I can interconnect. Fortunately we are using fiber so I see no distance limitations and there will be no loops in the network reducing any potential spanning tree issues.

Thanks,

Greg

1 ACCEPTED SOLUTION

Accepted Solutions
Gold

Re: Switch Design and sprawl

If you are going to be in a loop-free environment, then in theory there is no limit to the number of switches you can daisy-chain together. (Spanning-Tree would have imposed a default restriction of a network diameter of 7.)

However, in practice you will find that in moderate to heavy network traffic situations, users plugged into access switchports at the extreme edge of a daisy-chain will have a harder time getting to server resources located at the other end of the chain. Users who are closer to those servers will have an easier time of it. (More users are contending for bandwidth on the uplinks as you get closer to the servers.)

A way to minimize this contention, or at least limit it on the switch side, is to make your networks relatively "flat". That is, put in a Gigabit Ethernet aggregation or backbone switch, and plug as many 3524s directly into it as possible. Also, plug your servers directly into the center of your network as well. (Now your performance bottleneck will be the speed of your server connections, not the network.) In your scenario, this could be achieved by installing a 3550-12G or 3508G in the MDF.

My rule of thumb #1 is, daisy-chain no more than 96 users onto a single Gigabit uplink to the backbone switch.

There's no law that says you can't chain more together, such as a GigaStack cascade of 9 3548s in an IDF, riding a single fiber uplink back to the MDF. The limit is what level of performance your users will tolerate, versus what costs your accounting department or business office will bear in order for you to build an optimized network. (Finances usually win out, until the users start complaining and you can demonstrate that the cause is a less-than-optimum network.)

Rule #2, balance as best you can the Gigabit uplinks coming out of an MDF or IDF to the backbone switch. For example, if you have 4 3524 switches in an IDF, and you have more than one pair of fiber strands out of that closet back to the MDF, daisy-chain two switches each to a fiber uplink instead of three on one and one on the other.

Rule #3, set the copper 10/100 ports to Spanning-Tree PortFast enabled; and leave fiber or copper GBIC ports used for switch-to-switch connections as PortFast disabled.

Rule #4, configure fiber or copper GBIC ports used for switch-to-switch connections as Switchport mode trunk, and copper 10/100 ports to switchport mode access.

Rule #5, enable Spanning-Tree UplinkFast on all access switches in the IDFs and MDF, and disable it on aggregation/distribution switches in the MDF.

NOTE: Rules #3 and #5 optimize STP performance, even though you don't have loops in your network.

If cash is tight, and top-speed performance isn't a concern, you can turn your 5 3524s in the MDF into a "poor-man's chassis" by running GigaStack GBICs with redundant loopback cable on interface Gig 0/1 of each switch. Then, put fiber GBICs into Gig 0/2 and feed out to 5 IDFs. And in each IDF, create a similar GigaStack cascade, then use one fiber GBIC in each stack to reach back to the MDF.

Hope this helps.

1 REPLY
Gold

Re: Switch Design and sprawl

If you are going to be in a loop-free environment, then in theory there is no limit to the number of switches you can daisy-chain together. (Spanning-Tree would have imposed a default restriction of a network diameter of 7.)

However, in practice you will find that in moderate to heavy network traffic situations, users plugged into access switchports at the extreme edge of a daisy-chain will have a harder time getting to server resources located at the other end of the chain. Users who are closer to those servers will have an easier time of it. (More users are contending for bandwidth on the uplinks as you get closer to the servers.)

A way to minimize this contention, or at least limit it on the switch side, is to make your networks relatively "flat". That is, put in a Gigabit Ethernet aggregation or backbone switch, and plug as many 3524s directly into it as possible. Also, plug your servers directly into the center of your network as well. (Now your performance bottleneck will be the speed of your server connections, not the network.) In your scenario, this could be achieved by installing a 3550-12G or 3508G in the MDF.

My rule of thumb #1 is, daisy-chain no more than 96 users onto a single Gigabit uplink to the backbone switch.

There's no law that says you can't chain more together, such as a GigaStack cascade of 9 3548s in an IDF, riding a single fiber uplink back to the MDF. The limit is what level of performance your users will tolerate, versus what costs your accounting department or business office will bear in order for you to build an optimized network. (Finances usually win out, until the users start complaining and you can demonstrate that the cause is a less-than-optimum network.)

Rule #2, balance as best you can the Gigabit uplinks coming out of an MDF or IDF to the backbone switch. For example, if you have 4 3524 switches in an IDF, and you have more than one pair of fiber strands out of that closet back to the MDF, daisy-chain two switches each to a fiber uplink instead of three on one and one on the other.

Rule #3, set the copper 10/100 ports to Spanning-Tree PortFast enabled; and leave fiber or copper GBIC ports used for switch-to-switch connections as PortFast disabled.

Rule #4, configure fiber or copper GBIC ports used for switch-to-switch connections as Switchport mode trunk, and copper 10/100 ports to switchport mode access.

Rule #5, enable Spanning-Tree UplinkFast on all access switches in the IDFs and MDF, and disable it on aggregation/distribution switches in the MDF.

NOTE: Rules #3 and #5 optimize STP performance, even though you don't have loops in your network.

If cash is tight, and top-speed performance isn't a concern, you can turn your 5 3524s in the MDF into a "poor-man's chassis" by running GigaStack GBICs with redundant loopback cable on interface Gig 0/1 of each switch. Then, put fiber GBICs into Gig 0/2 and feed out to 5 IDFs. And in each IDF, create a similar GigaStack cascade, then use one fiber GBIC in each stack to reach back to the MDF.

Hope this helps.

140
Views
0
Helpful
1
Replies