When connecting to an Internet Exchange, you usually start with a small access size, depending on your needs. After setting up the initial connection, there will be a point when your traffic is growing and you will think about upgrading. But when exactly is a good time to upgrade your connection? We spoke to DE-CIX Academy’s Wolfgang Tremmel who explains why it is not a good idea to drive up a circuit too high before upgrading your connection to an Internet Exchange.
Stay in your comfort zone
Upgrades need to fulfil two goals: First, to have your connection in the “comfort zone”, and secondly, it should also be economically sound. What the comfort zone is, depends on your company and its willingness to take risks. Generally speaking, when you see continuous growth, you should start your upgrade process when a circuit utilization crosses 50%, so that you can have the upgrade in place once the utilization reaches 70%. 70 or 80% may sound very reasonable to a network engineer, but for the management level of your company, it might give the impression that money is being wasted. Why not drive a circuit up to 90% or even higher, to get most for your money? There are a number of reasons.
1. Having spare capacity for bursts of traffic
Even if traffic graphs look smooth, traffic itself is not. There are always bursts of traffic, which might not even be seen in the graphs. Causes can vary – a lot of large file transfers starting at the same time, re-transmissions because something went down for a short time, and so on. One thing is clear though – you need reserves in order to cover these bursts.
2. Having spare capacity for customer growth
You never know when the next customer will sign up and bring their traffic to your network. Therefore, is it good to have enough spare capacity to accommodate any new customer signups, and not turn away revenue just because you wanted to save couple of bucks by not having enough spare capacity. Obviously, it is not a good idea to overload the network with a new customer. For new customers, it is important that they like your network from the start, otherwise they will take their business elsewhere. It could also influence your existing customers, and you risk losing some of them, too. This would obviously resolve your capacity problem as well, but probably not the way you want to resolve it.
3. Having spare capacity in case of outages
This is the trickiest one. And always a toss-up between running a network as economically as possible and having enough redundancy. If you keep too much spare capacity, your costs are high. If you do not have enough and experience an outage, your customers complain because of network overload. As a good rule of thumb, your network must have enough spare capacity so that one outage (any single outage) is undetected by the customers. If you have multiple upstreams and peerings, designing them in a way that any of them can go down and the rest will take over the load – with still some capacity to spare (remember bursts…) – makes the most sense.
4. Sleeping through outages
In the case that an outage happens during the night and there is not enough spare capacity, the people in the management level are not the people to be woken up. The network engineer is. Maybe if, each time there is an outage, the engineer wakes up the budget holder as well, then upgrades will in future be ordered more quickly. Another way is to explain to the management level – with data – the business benefits of upgrading your access (see points 1-3).
Being prepared and planning ahead
As a real-life example, at DE-CIX, capacities in our network are regularly expanded on a long-term basis. We always plan about twelve months in advance, and we upgrade again as soon as 63% of the existing capacities are reached. The remaining 37% free capacity is needed to create redundancy and to ensure that we always have enough free capacity for the growing traffic needs of our customers.