A few years ago I was approached by a state organization to be a second set of eyes for a large purchase of Cisco routers and access points to connect all its branch offices to a new central hub. Most hardware line items were Cisco 2921s and 2951s and Aironet 1041s.
I asked how many end users would be at each site. I learned it would be about eight to ten, with many of them working remote in the field most days. I said the 2951s would probably be fine for a site with eight hundred to a thousand people if most of them are out in the field all day, but I’d be more comfortable with a 3900 series router for that many potential end-users.
Then I was corrected. Not eight hundred to a thousand. Eight to ten people.
I looked at the line items and immediately went to the total at the bottom: this organization was going to spend millions of dollars that it didn’t have to.
This is one example in a long string of disagreements I’ve had with solutions architects, sales people, and technical managers with regard to design decisions. I’m bothered by the level of overkill in network design and hardware selection, and I’ve become jaded by terms like “scalability.”
The business case for a design should drive network device selection, topology, and subsequently the cost. This presupposes a certain level of understanding of the network: traffic baselines, current port density, actual predicted growth (not completely hypothetical pie-in-the-sky growth), and the likely timeframe until the next hardware refresh.
Without this information, a design can be completely out of touch with how users consume the network, and costs can be outrageous. In my experience, this often leads to incorrect choice of hardware and designs that hint at fault tolerance but really only double the hardware and increase the costs.
These costs consume the resources needed to perform other necessary upgrades or augment staff and make the total cost of network access per end-user completely unreasonable.
I know some IT departments have a very difficult time finding money to complete projects or upgrade aging hardware, but in those occasions when an IT department has a decent budget or comes across some other source of funding, everyone needs to stop and take a very close look at reality.
- If your two year old WAN routers are doing everything they need to do just fine and their CPU utilization is 2%, then you really don’t need to replace them right now, and they certainly don’t need to be upgraded.
- If you need to replace all your old closet switches that provide connectivity to end-users who use their computers to check email and punch in and out with an online timeclock, the Catalyst 2960X with minimal options may be a much better option than the Catalyst 3850X.
- If you’re running two domain controllers, four application servers, a print server and maybe a couple file servers, you can probably cancel that meeting about Cisco ACI.
Think of the traffic baselines. Then think of the business use-case. Now think of how many more switches you could replace with the savings.
This applies to the actual topology as well. For example, it makes little sense to me to purchase millions of dollars in redundant hardware just to have it all sit idle because spanning tree has that entire section of the network in a blocking state.
Sure, there’s a design debate there: run Layer 3 to the closets or maybe focus new spending on redundant supervisor chassis switches. But my point doesn’t change: what’s the use-case? What are current traffic baselines? How much growth do we really expect in the next five years or so?
One last example. I saw a design a few years ago to increase bandwidth to a school district’s wireless LAN controllers to redundant 20Gbps port channels each. This was an effort to scale wireless capacity to accommodate BYOD devices on the network. The cost was great due to the hardware needed to meet this requirement.
I dug in a little bit – just a little bit mind you – and saw that their current total traffic to the WLCs was rarely more than 1Gbps during peak times. I learned they expected approximately 300 laptops and just as many cell phones to be added to the network.
I posited that the addition of a few hundred laptops and cell phones wouldn’t increase traffic even remotely close to what the design called for. The school district needed to upgrade their wireless – that’s true – but was this rightsizing the network?
I understand many factors that come into play when designing a network and choosing hardware. There are the politics of the C-level, the budget forecasts of the finance team, the skills of the IT staff, and the aggressiveness of the salespeople.
But for those of us in the trenches, for those of us designing networks, for those of us configuring hardware and writing proposals for the next fiscal cycle: it’s time we start to take rightsizing the network more seriously.
By the way, I ended up recommending to that state organization that they consider the Cisco 891w which had redundant WAN ports and the integrated wireless they needed. As of today, I believe they took my recommendation – saving taxpayers a few million smackers.