Are your servers past their 'Best by' date?

Just because something works, doesn't mean it's working well.

Checking product dates at the grocery store is common. "Sell by," "Best by," and "Use by" are a regular part of our lives as consumers. It's obvious when a perishable food such as milk goes beyond its expiration date. But the dates for other items, such as canned goods, can be less clear. Often a government requirement deems that product packaging must include this “best by” information rather than providing an absolute expiration date. Yet research shows that canned goods are often good for years, if not decades, past the label date.

Do servers have a "best by" date? Should they? Eventually, servers are no longer the best solution to their assigned task, but when they keep chugging along, "if it ain't broke, don't fix it" often becomes the replacement guideline. After all, why spend money to replace something that's working? Unfortunately, this perspective ignores the ongoing costs of maintaining operational efficiency.

Advances in server technology mean that a three-year-old server—considered state of the art when it was new—is significantly less capable than current technology. The advances in CPU, memory, and motherboard design mean that today's servers can better serve an environment where virtualization is commonplace. Newer equipment can more efficiently instantiate a larger number of virtual machines, and run them more effectively and at higher performance levels than was possible 36 months ago. And that comparison presumes that a company is on an aggressive three-year cycle for replacement of server hardware. In many cases, especially for lower priority services, servers are often left in operation for five or more years, as long as they are still running without need for replacement or repair.

The latest from Ponemon — 2016 Cost of Cyber Crime Study & the Risk of Business Innovation

Evolution in action

Let’s look at some simple basic hardware differences, starting with memory. While DDR3 RAM was the de facto performance standard for most of a decade, DDR4 RAM has advantages that translate beyond simple performance. While the clock rate of DDR4 memory starts where DDR3 topped out, it provides that greater performance at lower voltages, resulting in less power consumption. For a data center full of servers and memory, the difference in power consumption also has another happy result: the reduced cost of cooling.

The same power-performance metric is true of CPUs as well. According to an Intel brief on the Intel Xeon Processor E7-8800/4800 v4 product families, one of the major differences between the 2016 top-of-the-line E7-8890 v3 Xeon processor and the latest 2017 E7 v4 family of processors is that the newer generation can be almost 40 percent faster depending on workload. The newer CPU also has 25 percent more cores, albeit running at a slightly lower clock speed, but drawing the same amount of power. And this doesn’t even touch on the newer processors' ability to support larger amounts of memory and other additional features added to the latest generation CPU.

This alone means that a server configured almost identically to a system only a year older is capable of supporting more VMs in a smaller space and energy footprint, reducing the ongoing operational expense of the server or blade. And these metrics apply to server technology that changed over a year. Comparing a current cutting-edge server to one as little as three years old makes that older server look positively arthritic.

The impact of these server changes in the data center world is profound. Data center operators, both in the commercial space and the enterprise-class single business data center, used to have data center halls occupied by two or three dozen racks of servers. Now they are using the fully commissioned power for the hall in a single row of high-performance, high-density servers. They are supporting larger, more demanding workloads on fewer pieces of hardware and reaping the benefits in facility costs.

What is the actual cost?

When a server is operational for an extended period of time, the amortization schedule that applies to the hardware effectively makes the cost of the server zero. However, the acquisition cost is only a small part of the overall expense over the server’s lifetime. From the facilities perspective, there is the cost of space, power, and cooling to consider.

Obviously, the expense can vary greatly based on the type of server. Standalone, rack-mount, and blade servers have different requirements that can widely impact the aforementioned expenses, but IT can find itself overlooking the most significant expense: the cost of software.

Some types of software are licensed on a per-seat basis, such as typical knowledge worker applications and email. However, the most expensive software in the data center is usually licensed on a per-CPU or per-socket basis. When you are paying for a massive database cluster and the software license is based on a per-socket license, the ability to increase the performance of your environment by 25 percent per socket means you need to spend that much less on software licensing. And that is an immediate and positive impact on the bottom line.

Many IT shops have an unintentional disconnect between the people responsible for hardware and those for software. IT gets handed specifications for a project to deploy a new, large-scale application back end, with specific hardware needs, and it looks to meet that need by spending the least amount of the CapEx budget as possible. Meanwhile, the software people on the business side get told they need to buy a license that supports the hardware that IT selected. 

Three years later, the cost of the hardware has been amortized, but the annual software expense has remained the same or increased. When the idea of re-homing the application comes up, someone always puts up their hand and reminds everyone else that the application is stable and meeting current business needs, taking the “If it ain’t broke, don’t fix it” approach. And in many minds, the CapEx and OpEx budgets are treated as isolated processes, completely overlooking the potentially huge savings in annual licensing fees that upgrading even relatively new servers to the cutting edge can bring.

Because it is the financial aspect of letting those old servers chug along and not necessarily a technical issue, someone in the organization who understands both the technology and business parts of the equation needs to step up and look at the actual operational expenses of keeping those existing servers in place.

It is an imperative to examine the facilities costs for space, power, and cooling, and the licensing costs of the applications running on those servers. Only by making a direct comparison between ongoing operation costs, acquisition costs, and the OpEx savings of moving to current technology can a business operate at its highest level of efficiency.

Lessons for leaders

  • "If it works, don't fix it" may sound like reasonable advice, but it doesn't apply when newer technology brings far more benefits than the old equipment.
  • As you consider server upgrades, look to license cost savings and environmental improvements.

Related link:

Serverless and the rise of the non-developer developer