Business details
Registered company no. 11869849
VAT number 317720513
In this story, there will be those that will be thinking and probably saying that I am wrong. Just thought I would mention that up front! I have an opinion, as do you. Mine led to one conclusion at the time that I probably wouldn’t do today but the technologies involved are far more mature and at the time, the approach worked flawlessly.
To frame this up, we are going to talk about QoS (quality of service). We are however back in around 2003 maybe, early days of this and early days of reliable VoIP (voice over IP).
I am in this story responsible for the technology and configuration of the LAN elements of a very large multi-site rollout of a new LAN, WAN and VoIP capability. Making any VoIP solution work is about end-to-end consistency of packet marking to a standardised QoS model, so bear that in mind.
Whilst I am building up a standard set of deployment templates for each of the different switch types in the rollout, people keep asking me how I am going to handle the QoS. This is bugging me because, quite frankly, I don’t know. I have near zero QoS experience but understand the concepts and the principles yet the confusing array of QoS stuff really threw me. Looking back, given the complexity level, I am not surprised I was finding it confusing.
To make it even more confusing, the chassis based switches had different input and output queue setups on a per line card basis. This meant to me that to truly do end to end QoS, the amount of templates I was going to need and the complexity levels within those templates was just escalating rapidly. It was really confusing me.
Then I had an epiphany. We were putting in top class devices and there were not any points of congestion then actually deploying a QoS setup on a huge multi–terabit LAN switch was a daft thing to do. You were probably going to slow down the flow of packets more by deploying QoS than by leaving the line speed switching capability to just do what it does best. We could then identify the traffic just prior to hitting a point of congestion and mark it up at that point (i.e. the WAN router).
Thus we started an experiment to prove me wrong. End to end QoS for VoIP without deploying it in the LAN. Configurations dropped from many hundreds of lines, with some extreme complexity to a bare minimum.
Oddly, or maybe not so oddly when you really think about it, it worked. For many, many years. Eventually we did go back and place standardised QoS configuration on the equipment but it was many years later and not because of any problems but because we had given ourselves the time to tackle this problem in a very thorough and robustly tested way.
Learning point for everyone:
Aim for simplicity. There is often a grand desire in the technical world to deploy the latest and greatest new feature that drives a software or hardware upgrade that will do better IT things by doing something exciting for geeky types.When you see this, stop for a second. Think really hard, what does your company do? Does this technical thing allow you to make more or better widgets or reduce the cost of creating those widgets? If not then maybe you don’t need to spend all that effort learning this new thing that will add complexity and overhead into your operational teams.
The LAN space is full of this kind of thing, really ask yourself if just a fast switch would do enough. Is 802.1x really a brilliant security feature or is it an overhead to your business? That may depend vastly on the physical location of your buildings and the types of physical access control you have along with the predisposition to allow visitors in.
QoS remains a good example, as the complexity became more and more apparent to those using it Cisco, as an example, developed auto-QoS – acknowledging that the user interface to this technology needed simplification. This we will see more and more of as times move on, the initial complexity of something new having layers of what is essentially new user experience then added in order to simplify.
There are so many businesses around that are just creating problems for the customer. The proliferation of AI and MI, alongside data analytics is a current great example. If you as a business find new great insight in someone’s data they then immediately have the problem of who will look at those results and what will they do with that new information.
A wise man once said to me, if you tell me about a problem, I have to deal with it. If it was found out I knew about it and ignored it that would be my doom. Please don’t’ give me fresh insight, I don’t have the budgets or people to handle it. In his particular case in a highly regulated cost constrained industry, it made sense. It doesn't in all cases.
Registered company no. 11869849
VAT number 317720513