Server Fault is a question and answer site for system and network administrators. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I was told that a network's graph connectivity is a "dangerous" way to judge reliability. Why is that?

In my mind, a complete graph (the most connected graph) is the most reliable network possible.

For those who dosen't know what I mean by "graph connectivity" or "k-connectivity", see this link :

Thank you.

share|improve this question
up vote 4 down vote accepted

It's a dangerous for a real network because it's applying a narrow metric when you need to take a holistic approach.

In graph theory, reliability is best when you can remove the most paths on the graph, and still retain a valid path between all vertices. Hence, a network where each vertex connects to all others with a dedicated paths is the most reliable.

In real networks, that's just one of many factors. Reliability is best when your servers and workstations can happily communicate at acceptable speeds come rain or shine and that's determined by a lot more than simply 'how many links to other things do I have?'

In real networks, you must take into account:

  • Protocols
  • Geographical Location
  • Finance
  • Human Fallibility
  • Compatibility
  • Interference
  • Business/User/Vendor/Stakeholder expectations

Any one of the factors above (and probably a whole bunch more) will impact the reliability of a real network. So you can see that a holistic view must be employed to properly gauge a real network's reliability.

Applying computer science directly to IT scenarios tends to be problematic because real-world factors are rarely considered in detail

Here's a few real-world examples I've encountered where the reliability of a network couldn't be measured by k-connectivity:

  • The computer infrastructure supporting a 24/7 steel mill operation. We'd regularly encounter reliability problems due to electrical/magnetic interference on the network links. For this reason, many cat5 runs had to be converted to optical links. It wouldn't have mattered if we had 1 or 100 cat5s for the runs that were near active furnaces. You'd still see regular network drop-outs due to unpredictable EMF.
  • A network mix of IPX and IP running over Netgear, Cisco, HP, 3COM and unbranded network equipment. Connections would drop. IPX traffic would mess with measurement and flow of IP traffic. Systems couldn't be universally monitored and we had interruptions in service simply due to being unable to detect and repair problems quickly enough.
share|improve this answer
Thank you for you answer Chris. If I understand, connectivity is only one of many factors to consider when evaluating a network reliability. – user57366 Oct 17 '10 at 18:42

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.