The downside of redundancy
Redundancy sounds like a no-brainer: more certainty, less risk. But this concept also comes with its own technical and practical challenges.
The biggest pitfall? Geographic distribution. As soon as your infrastructure is spread across multiple physical data centers, for example in Brussels and Antwerp, latency inevitably arises. This phenomenon occurs because data now has to travel between physical locations, via networks that are subject to distance and therefore introduce delay.
Such a setup between these two cities causes on average about 4.8 ms TCP latency. Not catastrophic on paper, but in reality it is noticeable: slower load times, delayed response times, or small lags in transactions that can add up to a significant difference in performance.
So how do you solve this latency? The trick lies in smart architecture. Important components are often duplicated per location, so most traffic stays local and only replication or synchronization tasks cross this distance. Techniques such as caching, load balancing and asynchronous replication help reduce the impact of latency significantly.
And what about data? As soon as you need to persistently store data, for example in a database, extra challenges come into play here as well. When a user saves something, such as an order in a webshop, that same write operation must end up in both locations, so that the data remains identical everywhere, or in other words, “in sync.” This can be done with sync writes (safe but slower) or via async replication (fast but with the risk that the replica is not fully up to date). In the case of a failure during async replication, this may mean that your recently stored data is lost. In short, both options impact the performance and complexity of your setup.
And that is the paradox of redundancy: you choose it to make your hosting environment more resilient against downtime or outages, but without the right considerations it can negatively affect the user experience. At Level27 we combine redundancy with a network and infrastructure design that minimizes latency, so that you get the certainty of distribution without any noticeable performance losses.