Servers have always powered businesses and also formed the backbone behind how they run and operate. There has been a marked increase in reliance on servers, which are now being used to do things they never used to do before.
Looking at an industry where servers play an integral role, data centre administrators can be said to be the most advanced and complex folks in the IT world. Workloads on servers, and the servers themselves are getting more complex now especially post-virtualisation. As a result, data centre administrators are faced with the constant challenge of fitting the load to the infrastructure. Is it an intensive task which needs high performance compute-style servers or a financial transaction which needs speedy processing and a low-latency network connection?
Exponential demand means that increased performance is key and these data centre professionals face the constant challenge of getting new levels of performance out of servers that have existed for years in their data centres. The key question here for data centre professionals is how do you scale your data centre’s performance?
Scale up – vertical
This process involves installing more high performance components in your existing servers, usually in the order of processors, memory, and storage (by using SSDs to replace hard drives). Sometimes the order flips and we might lead with memory rather than processors; it all depends on the configuration of existing servers and what we’re working with.
If what we’re trying to improve is an older DDR3 server, the first thing to tackle would likely be the CPUs if the server was using Newhalem or Westmere processors. There’s a lot that goes into this though, and we might start with memory if the DRAM was sparsely populated. It all depends. Key questions to ask would be, what’s your workload? And which components are going to have the biggest impact on what you do?
Scale out – horizontal
This can be seen as a stair step process. Typically when scaling out or up, the first step is getting as much as possible out of existing servers. Once this is done, the only way is to go is horizontally, adding more servers. Scaling out is definitely more expensive, as it involves more physical servers – and thus more software licenses. Since software licenses usually cost far more than server hardware and hit your budget every single year, they’re one of the most important cost factors when considering whether to scale out.
Additional factors such as higher power and cooling costs all add up when you’re working with a greater number of servers, and these are just a few of the costs associated when scaling out, on top of the issue of having physical space in your data centre.
The decision making process
Determining which approach to use is all about knowing exactly where your organisation is in the expansion process and which scaling decision would be the best fit for where you’re at – and how you see your workload growing.
IT decision makers who need to take this first step in deciding between scaling up and upgrading servers – or scaling out and adding more servers – need to bear all these considerations in mind. Both routes have their advantages. It goes without saying that typically it’s more efficient to first maximise the performance potential of your existing infrastructure and servers by scaling up. Once a performance ceiling is reached then it’s time to scale out to further meet demand.
The challenge for IT decision makers is in identifying exactly where their organisation sits in that process, and therefore which route is the most effective way to go.