LAN Sizing
From Wikipractice
Contents |
Introduction
The initial sizing of a LAN is more complex that it seems at the beginning. It is more an art than a science, as it is linked to the experience of the designer.
The present document gives some rules that have been successfully applied in different industry branches. However, these rules does not prevent a proper performance measurement and performance analysis on an ongoing way.
Bandwidth
On LAN technologies, the usual bandwidth are 10Mbps, 100Mbps, 1000Mbps and starting now in some cases, 10000Mbps. The different bandwidth profiles and their usages are described in the table below.
| Description | Bandwidth |
| User workstation | 10 or 100Mbps |
| IP Phone | 100Mbps |
| Server | 10, 100 or 1000Mpbs |
| Connection to WAN router | 10, 100 or 1000Mpbs |
| Connection between switches | 1000Mbps or an aggregation of n*1000Mbps |
| Now starting with 10000Mpbs |
Bandwidth for IP Telephony
Bandwidth used by a telephony channel depends on the codec used to compress the voice traffic. In the worst case, we can consider that IP telephony use 10Kbps per channel.
Amount of ports
When designing or upgrading a LAN infrastructure, it is necessary to be prepared for growing. Experience shown that a 30% free port port margin is a good practice (1.3*used connections). As an example, is a LAN needs 16 connections, it is better to install a 24 port switch.
When calculating the amount of port, don't take the amount of users into consideration, but the amount of necessary connections. Depending on the industry, users may needs ore that one port each.
Overbooking
The network has to be designed to provide a reasonable capacity to its users. It is an optimized compromise between costs and bandwidth.
As we know that not every user will consume all their available bandwidth at the same time, we use overbooking factor to install less bandwidth that the total access connectivity. These numbers are coming from the experience acquired in designing and operating infrastructure.
A bandwidth overbooking factor is a relation between the bandwidth available in two different parts of a device of the network. Overbooking factors are defined in table below:
| Where | Factor |
| Between access and distribution | 20 |
| Between Distribution and core | 4 |
Overbooking between access and distribution
The overbooking factor between the access layer and the distribution layer is the ration between the sum of the bandwidth of the access ports and the sum of the bandwidth of the liks to the distribution layer.
The overbooking factor must be calculated withe the active links. As an example, if link X is the production one and link Y the backup one, you can not use the sum X + Y to calculate the overbooking factor. Only X can be taken in account..
Example of calculation
Lets take the example of a 48 port access switch:
- Each access port is 100BaseT
- Link X and Y are 1000BaseFX
- Link X is the production one
- Link Y is the backup one.
The total access bandwidth is 48*100Mbps= 4.8Gbps. Overbooking factor is then:
Which is less that 20.
Between distribution and core
The overbooking factor between the access layer and the distribution layer is the ration between the sum of the bandwidth of the distribution ports (A) and the sum of the bandwidth of the links to the core layer (B).
Usually, we do use layer three routing between Distribution and Core, with load balancing. We take all links between Distribution and Core, but only the active links between Access and distribution.
Example of calculation
Lets take two distribution switches each with 4 active 1000BaseFX links, connected to the core via 4 redundant 1000BaseFX links.
Which is less than 4.
It the overbooking factor is greater that 4, the solution is to add links between Core and Distribution. The best way is to create trunks. Take care of being symmetrical.
Overbooking with a stack of access switch
Stack of switches are usually used at the access layer. Stacks are build in connection several physical devices to create a larger logical one through a special connection, most of the time proprietary. The stack behaves like one unique device, and the connection to the access layer is usually done via one or two members of the stack, not all.
The overbooking factor of a stacked switch has to take in account all access port.
Example of calculation
Lets take the example of the following stack configuration:
- 5 switches stacked together
- Each switch has 48 100BaseT ports
- One primary 1000BaseFX link to the distribution layer
- One backup link to the distribution layer
5*48*100Mbps = 24Gbps.
Which is greater than 20. Therefore it is necessary to increase the bandwidth between Access and Distribution.
Subnet size
A subnet is a layer 3 network division, consisting of a range of IP addresses. Subnets are classified in 3 main categories:
- A, 16'777'216 IP addresses
- B, 65'536 IP addresses
- C, 256 IP addresses
A subnets are usually linked to VLAN, which are layer two network division.
Best practice in network design strongly advise to not overpass 254 devices per per VLAN: this corresponds to a C class in IP layer. There are two main reasons for this:
- Limiting broadcast domain.
- Limiting broadcast storms.
Broadcasts consumes bandwidts and CPU resources on evera devices which receive them. As an example, I saw a flat network with 44096 addresses within one of my customers: the broadcast level was aroung 2Mbps, and the PC where conencted at 10Mbps: 20% of the bandwidth was just waisted for nothing.
