“One-to-one” migration to cloud is missing the point

If you move on-premises servers to the cloud in their original CPU, RAM or disk configuration, you may end up paying more than you need to. In the cloud, servers are sized differently because fewer resources are needed to achieve the same computing power.

Most enterprise servers are already over-provisioned, with the typical average CPU utilization coming in well below five percent, to allow for load peaks and future growth. In a cloud environment, resources can be added or removed instantly using a web interface or API, so there is no need to buy reserve capacity. Why pay for air? It is true, of course, that adding a processor or memory requires restarting your operating system, but that only takes a few minutes. When an IT service is horizontally scaled on a load-balanced group of servers, an entire server can be added or removed without causing any downtime. DT’s Open Telekom Cloud does this automatically if your server load rises above or falls below a certain level.

To obtain the same computing power in the cloud as on your physical infrastructure, a much smaller virtual machine will do the job. This is because a cloud server can perform at higher loads: We recommend maintaining it at between 30 and 70 percent of its total capacity. Another reason is that cloud providers renew hardware on a regular basis, so their more recent CPUs may perform better than what you have at home.

Let’s look at an example: Your 4 vCPU server has a 17-percent average load on an E5-2620 V2 processor with a Passmark benchmark of 1299. You look for a suitable replacement in DT’s Open Telekom Cloud, which uses E5-2658A V3 CPU benchmarked at 1682. Because of the low on-premises utilization and higher cloud benchmark, you can replace your quad core server with a single (!) core cloud machine, and the resulting estimated load would be: 4 x 20% x 1299 / 1805 = 58%. This is fine, because if the load rises in the future, you can simply switch to a 2 vCPU server.

Some cloud providers allow you to create multiple virtual machines or containers inside a single virtual server, in a setup resembling the Russian matryoshka nesting doll. This can be an economical way of running a large number of distinct, low workloads. Further savings can be achieved by buying reserved capacity, which usually carries a significant discount, and entering into a binding contract for at least one year. You should also ask your provider about transferring your existing on-premises OS licenses to the cloud.

Cloud computing can save you money – but you do have to learn to map your existing workload to the cloud more intelligently than with the common “direct match” strategy.