Scalability Flexibility and Modularity in Network Design

Scalability

Ability to perform changes without changing the entire architecture / design.

There are 2 scalability approaches for the network designs.

1. Scale Up
2. Scale Out

Scale Up

Increase the existing system resources without adding a new system.

Scale Out

Having one physical router which can add multiple line cards later on is an example for Scale Up type of scalable solution while having 2 physical routers is considered as the Scale Out solution.

image ref: medium.com

Flexibility

Ability to adapt to business changes.

Modularity

Ability to divide by functions or policy boundaries.
There are 3 modularity approaches which provides flexility to a design.

1. Choosing the physical topology

Some topologies such as hierarchical or leaf and spine are easier to work with adding modules than fully meshed etc.

2. Splitting Functions or Geographies

Separating campus, branches, data center, internet edge etc or security policy boundaries make the design easier to upgrade, manage etc.

3. Break into smaller pieces

Creating smaller fault domains so that a failure on a part of the network doesn’t impact whole system. Not extending the spanning tree domain between data centers is an example.

Modular design allows different modules to be managed by different teams. Network Team, Firewall team, Data center team etc in Enterprise Networks or Core network and Access network in Service Provider networks are examples.

Also modular designs can reduce the configuration overhead, template based configuration in SD WAN is an example.

Packet Delivery Parameters and Reliability in Network Design

There are 3 packet delivery parameters.

1. Delay / Latency

Time which a legitimate packet takes to travel from source to destination. 

2. Jitter

Consistency of delay / latency

3. Packet Loss / Drop Ratio

Fraction of packets sent by the source but not received by the destination.

General accepted best practices for the Delay, Jitter and Packet Loss ratio has been defined per type of applications. For example, one way Delay for VoIP (mouth to ear delay) should be less than 150ms, Jitter should be less than 30ms and PLR should be below 1%.

Reliability

Delivering the legitimate packets from source to destination within a reasonable delay / latency which is defined based on the application type and architecture.

Reliability is often mentioned in choosing links. As an example, if you have to utilize a mix of reliable and unreliable links, best practice is to carry VoIP traffic over the reliable links which don’t have packet loss and latency, and use the cheaper unreliable links such as internet to transport packet loss tolerant application traffic.

But reliability should be considered for everything in the path, including links, devices such as switches, routers, firewalls, application delivery controllers, servers, storage systems etc. Even the hardware onboards should be reliable too. That’s why vendors bother with ASICs, Quantum Flow processors etc in their marketing propagandas.