The Cost of Networking Has Not Declined

One of the common taglines parroted by SDN aficionados goes along the lines of “The cost to acquire and manage server and storage architectures has declined over time while networking stays stubbornly expensive.” (I took it straight from an anonymous blog comment).

Let’s see how well it matches reality.

The cost to acquire networking has not declined. I call bullshit. A 64 kbps leased line was 2-3 orders of magnitude more expensive 25 years ago than a 100 Mbps Ethernet circuit is today, for a total of 5+ orders of magnitude, which is almost exactly what the most common variant of Moore’s law (doubling every 18 months) would predict.

On the other hand, it seems that the high-end residential Internet bandwidth grows at approximately 50% year-over-year.

In any case, networking (like any other IT hardware technology) is getting ridiculously cheaper over time.

The cost to acquire server architectures has declined. Maybe. However, the servers, laptops, smartphones or tablets I’m buying are not getting cheaper, they’re just getting more powerful (and the bloatware running on them quickly consumes the extra resources). The same thing is happening in networking. You can get dirt-cheap 1 Gbps switches today (16 port switch costs less than $100 on Amazon) – compare that to what we were paying for them a decade ago.

The cost to manage server and storage architecture has declined. Not everywhere. Storage arrays are still managed with clickety-click UIs, and many enterprises still haven’t figured out what server automation is all about. I’ve seen plenty of enterprise environments where deploying a new VM takes weeks (because so many teams get involved)-

However, I do agree that more people use automation to manage servers than to manage networks.

Networking remains expensive to manage. This is actually a paraphrase of what I think the comment means. Not true – there are numerous organizations (from traditional ISPs and even some traditional enterprises to startups and large web properties) automating their network configuration and management. Some of them have been doing that for decades.

In a more traditional environment, nobody wants to automate a networking device because of excessively large blast radius (listen to this podcast with Jeremy Schulman for more details). Cloud-scale technologies might not apply to your network.

Also, there’s a reason designing cars differs a bit from writing software ;)

Networking can be cheap. Use open-source software on a Linux box instead of routers. Ah, you want to have something that will survive for 5 years in a dusty rack without air-conditioning? Well, you can buy an expensive ruggedized server ;)

Ah, you want support. Well, use a third-tier company like Mikrotik. Not good enough? Maybe what you really want are all the bells and whistles you get from a major vendor at the price of the underlying hardware. Can we stop this discussion, so I can go back to real world, and you can wander back to your pipe dream?


  1. More expensive? Really? £11k pa will get me 100mbps point to point Ethernet circuit these days, I remember 17 years ago having a 2mbps circuit costing somewhere in the region of £16-17k pa.

    Things used to be a LOT more expensive, I remember when we were looking at a pair of Cat6509s for the datacentre back in 2002, and it made my eyes water at the price.
  2. Did you mean miKrotik ( instead miCrotik? If yes I must to say: they have a horrible support. :)
    1. Yep, that one. Fixed.

      "... they have a horrible support" << Cheap, Fast, Good. You get at most two out of three. Worst case you get none ;)
  3. You can get a 10 gig point-to-point Ethernet wave that crosses the Atlantic ocean for less ~10k per month. Yes, networking has gotten much cheaper. Unfortunately, we still have to deal with the same clueless, network stupid executives who spend all their money on magic beans (i.e. cloud orchestration suites), but are the first to complain with the poor performance of their network.
  4. To be fair sfps are still expensive, but the main reason could be due to politics. In most orgs I have worked in, I've never seen anyone in a senior leadership role with a networking backgound, unless it was someone who had a legacy voice background. Even in IT, most leadership comes from the dev, database or PM side. So there is no one in authority to refute or at least look skeptically at the claims. Networking now, is for sure cheaper than the days of ISDN, frame relay, ATM, Token Ring, FDDI, Packet over SONET.
    1. Don't buy first party optics and you'll save a ton of money. Cisco or Juniper branded optics are good for just one thing - getting tech support to fix your problem.
  5. servers have moved from pets to cattles, but network devices are still pets ... and pets are more expensive than cattles
    1. And when was the last time you deployed a VM in an enterprise environment? Someone mentioned "time to deploy a VM" as the biggest problem they're facing in their DC.

      Just because clouderati talk about pets and cattle doesn't mean that it's true in typical enterprises. Likewise I could give you examples of network devices being treated like cattle (for example, every cable modem :D).
  6. Good points. Have you considered the costs within a limited world view though? That is, if you included the entire world of costs, relative to productivity in the past, what would the actual status of your question be? For example there is no doubt that technology increases the productive power of an economy. And this increase is relative, all else being equal. So how much did a 64k link technology increase productivity in the 80's compared to a 10G one today? I dont have a scientifically priced answer. But I remember working on 64k multiplexors and the price improvement 'felt' as good as a 10G today. The thing which I forecast is tipping the balance against your case is the need for more security the more Internet is used. And obviating the security leviathan with some kind of radical new technology, much as the Internet did at inception is surely the objective of the likes of you and I, rather than to relax and pretend 'everything is fine'
Add comment