Basic Understanding of Over-Subscription

Over-subscription in IT, specifically in network design, refers to the practice of selling or provisioning more resources than are physically available, based on the assumption that not all users will require the maximum available resources simultaneously. This is a common practice in the networking industry and can help reduce costs and optimize resources.

When it comes to the Core, Distribution, and Access layers of the network, over-subscription is typically applied as follows:

1. Access Layer: This is the first and most basic level of the network. It is the point where devices like computers, printers, and IP phones connect to the network. Here, the over-subscription rate can be quite high, as not every device will be sending or receiving data at the same time. Typical ratios can be 20:1 or even higher. If you’re working with ports of different speeds, you’ll want to ensure you have enough bandwidth to handle peak demand. For example, if you have 20 devices with 1G ports connecting to a switch, you’ll need at least a 10G uplink to handle peak demand and maintain the 20:1 over-subscription ratio.

2. Distribution Layer: This layer aggregates data from the access layer switches and routes it through the network. The over-subscription rate here will typically be lower than at the access layer, because this layer is handling traffic from multiple access layer switches at once. A typical ratio might be 4:1 or 5:1. For instance, if you have four 10G access switches connecting to a distribution switch, you’ll need at least a 40G uplink to the core to maintain the 4:1 over-subscription ratio.

3. Core Layer: This is the backbone of the network, routing traffic between different parts of the network and to external networks. It should have the lowest over-subscription ratio, as it is responsible for handling all the network’s traffic. In many modern networks, the aim is to have no over-subscription at the core layer (a 1:1 ratio) to ensure the highest possible performance. However, depending on the specifics of your network and its bandwidth demands, a small degree of over-subscription may be acceptable.

These ratios are guidelines rather than strict rules. The optimal over-subscription ratio will depend on the specific use case and the expected network traffic. Factors to consider include the types of devices connected to the network, the applications being used, and the tolerance for latency and congestion. It’s also important to monitor the network and adjust these ratios as needed to ensure the network can handle the demand and provide a good level of service.

Oversubscribing is basically when you go from many ports to fewer ports. So for example, a switch has 48 ports on it but only two uplinks. You can see right away there might be a problem if all 48 ports are being used and the traffic is being shoved down two ports. This brings you to a 2.4:1 ratio assuming the access ports are 1G and the uplinks are 10G which is typical. This is oversubscribing. When you oversubscribe you get congestion. When this occurs, QoS is required to protect important traffic.

 

LEAF AND SPINE

In a Leaf-Spine network architecture, the traditional core, distribution, and access layers are replaced with a two-layer design consisting of Leaf switches (access layer) and Spine switches (core layer). This architecture is popular in data center environments due to its high levels of redundancy and scalability.

The concept of oversubscription still applies in this context, but the way it’s implemented is different.

Leaf Switches: These are the access layer in this architecture, connecting to servers and other endpoints. As in the traditional three-tier architecture, the oversubscription rate at this layer can be higher because not all devices will be transmitting at their maximum rate all the time. However, in a data center environment, the ratio may be lower than in a typical enterprise network due to the higher performance requirements. Ratios like 3:1 or 2:1 might be more common.

Spine Switches: These form the core of the network in this architecture. All leaf switches are interconnected via the spine switches, with no direct leaf-to-leaf connections. To ensure optimal performance, it is typically recommended to have no oversubscription at this layer (a 1:1 ratio). This means that the total bandwidth available from the leaf switches should not exceed the total bandwidth available at the spine switches.

For example, if you have 10 leaf switches, each with two 40G uplinks, you would need sufficient spine switches to accommodate 800G of total uplink bandwidth without oversubscription.

Again, these are just guidelines, and the specific needs of your network may require different ratios. Factors like the expected traffic patterns and the performance requirements of your applications will play a role in determining the optimal oversubscription ratio.

 

CALCULATIONS

Oversubscription is calculated by comparing the total potential demand for network resources from downstream devices (like computers, servers, or lower-tier switches) with the total supply of those resources from upstream devices (like higher-tier switches or routers). Here’s a step-by-step guide to how you can do it:

1. Calculate Total Downstream Demand: Add up the total capacity of all downstream ports. For instance, if you have 24 devices each with a 1G port connected to an access switch, the total downstream demand is 24G.

2. Calculate Total Upstream Supply: Add up the total capacity of all upstream ports. For instance, if the access switch has a 10G uplink to a distribution switch, the total upstream supply is 10G.

3. Calculate the Oversubscription Ratio: Divide the total downstream demand by the total upstream supply. In this example, the oversubscription ratio would be 24G / 10G = 2.4:1.

You would perform similar calculations at each layer of the network, using the output from the lower layer as the input for the next layer up.

Remember, these calculations assume that all devices could potentially be using their maximum bandwidth at the same time. In reality, this is unlikely to be the case, so a certain degree of oversubscription is generally acceptable. However, you should monitor your network to ensure that oversubscription doesn’t lead to performance issues, and adjust your network design as necessary.

In a Leaf-Spine network, the calculation would be similar. For example, if you have 10 leaf switches each with two 40G uplinks, your total downstream demand is 800G. If you have enough spine switches to provide 800G of uplink capacity, then your oversubscription ratio is 1:1.

Also, bear in mind that this oversubscription ratio is typically calculated on paper during the design phase. Network monitoring and traffic analysis tools can then be used to verify these ratios in the live network, and adjustments can be made if necessary.