In my view, the most common architectural flaw made by network engineers is that the data centre has a single network. I believe that the correct perspective is that any “network” is a “network of networks”.
Lets run them down – DMZ, Internet, DCI, Production, Pre-Production, Dev/Test, Storage are a just a few of the different and separate. I quickly drew up a rough outline of some of the data networks that are in every data centre today:
All of these networks are connected (more correctly, interconnected) to form a network of networks. All endpoints are able to communicate except where limited by security policy in firewalled control points.
For some customers, some of these networks are futher ‘sub-networked’. For example, a legacy three-tier switch network that copes with the limitations of spanning-tree is often described as three network layers “Core/Distribution/Access” with each layers having its own properties.
Each of these network have differing hardware and software design criteria.
- Core Network: port density, high speed, limited features, L2 propagation
- DMZ Network: weird features, limited port count.
- Data Centre Interconnect: Support for L2 extension, optical support, granular traffic control (QOS/HQOS/DQOS etc)
- and so on
When considering strategic investment planning over five or ten year cycle, the “network of networks” is a valuable tool. Its more practical to describe smaller systems and then prepare a progressive strategy paper that solves smaller problems in a manageable and practical way.
Sure. This isn’t as ‘heroic‘ as designing a Titanic-sized network with huge dollar values and gaggles of sales grunts desperate to get a sniff of the deal. Missing out on a few free lunches with unlikeable people should balanced against an improved personal life when each network upgrade is small, management and peaceful.
Small & Focussed
Recognition of the “network of networks” as a design principle leads to new ways to allocate budget and improve services incrementally.
- Treat each network as a separate design space with challenges and criteria unique to that network.
- Have a master architecture to connect each of these networks together in stable and reliable way. e.g. L3 routing, IP addressing strategy, DNS/DHCP conventions.
- Upgrade each network according to business need.
- Reduce the scope of your projects to focus on each sub-network instead of attempting to solve ‘all’ the problems
- By products/solutions/platforms that are specific to each network.
- Introduce new technologies in small amounts for each network area so learning and adaptation can be achieved through organic continuous change
The EtherealMind View
Everything is connected and might lead to a cognitive bias where you see the network as a single system. This perception leads to “lift & shift” or “forklift” solutions that require complete replacement of joined systems. For most cases, it is better business for your employer to have a continuous rotation of systems than the current approach of buying once-in-a-decade network upgrades that suits a vendors business model.