New Rules For Refreshing Server Hardware

[#1]

Barring un-repairable failures, server, computer hardware can easily run five, even ten years. But, as with old automobiles, refrigerators, and cell phones, Ben Franklin's advice, "'Use it up, wear it out," may not be the most cost-effective – or productivity-effective – strategy.

Historically, companies "refreshed" —bought new hardware – based on accounting amortization timeframes or when new software needed more "oomph" (such as more raw CPU performance or more RAM). Some companies might assess whether, given that hardware typically represented 10-15% of TCO (Total Cost of Ownership), refreshing after three or four years was more cost-effective than eking that extra year or two out of the initial purchase investment.

Companies still have to refresh their computers. But it's no longer quite that simple.

Here's a look at what is – and isn't – driving today's server hardware refreshes, and how much shorter – or longer – they might be.

"The standard lifetime is five years for capital expense amortization of physical servers where I work," says Ed Frankenberry, a systems architect for a financial investment firm. "That's not to say that we don't keep any servers longer than that – just that they're typically obsolete by then."

Today, server hardware refreshes are often based on hitting one or more "trigger point" — specific events or other reasons – rather than a simple one-size-fits-all IT policy. In some cases, the time frame may not change; but even then, the reasons may.

Virtualize and consolidate: The first and obvious reason to refresh server hardware is to consolidate multiple server applications onto fewer (and generally, newer) machines using virtualization.

Virtualize-and-consolidate doesn't spell the end of hardware refresh, of course. But it does change the when and why of refresh triggers. IT may be able to add more capacity by replacing CPUs, adding memory, or sticking in additional blade or rack-mount servers.

Once you virtualize your server applications, says John Sloan, lead analyst at Info-Tech Research Group, "The trigger to purchase new hardware is about capacity planning, and maintenance of capacity to run all your servers, versus replacement of individual servers."

A lot of applications are still not virtualized onto "consolidated" hardware. According a recent mostly-enterprise survey, says Mark Bowker, senior analyst at Enterprise Strategy Group, According to Bowker, 58% of organizations have virtualized less than one-third of their servers, and 59% have not yet virtualized any mission-critical applications.

Control utility bills: A few years ago, the three-year cost for a server's power and cooling were roughly the same as the hardware's purchase price. But servers continue to get more energy efficient, to the point where utility savings can justify shortening the refresh cycle.

"In 2003, our first AMD Opteron processor was a single core in a single socket, and used 90 watts of power," says Brent Kerby, senior product manager for the AMD Opteron processor. "Today, for 85 watts you can get a twelve-core processor – and you get a lot of Instructions Per Clock (IPC) improvements for each core, along with power-savings features."

Performance per watt has improved dramatically," says Nathan Brookwood, research fellow at Insight 64, a market research and consulting organization. "It's often worthwhile to move to the newer technology just to glean the power savings. People have done ROI calculations showing you can recoup the cost of hardware in 12 to 18 months just through power savings, even for the same performance."

"You can almost get to the point where you can't afford to keep the servers you have been running, compared to the new systems," says John Gromala director, Industry Standard Server Product Marketing at HP. "Older systems weren't designed with this level of power efficiency."

For example, says Gromala, "In terms of power, HP's Generation 6 servers and new Generation 7 servers provide big improvements in energy usage of servers. Combined with better performance from the latest processors, we see companies whose previous hardware was three to five years old paying off their investment sometimes in as little as thirty days."

Cap power, cooling and space consumption: Even when the power and cooling costs aren't a trigger, the reality of maximum capacity can be. Data centers have a finite amount of space, cooling, and utility power. A hardware refresh, especially in the first rounds of consolidation-oriented, can mean the difference between maxing out your data center's power, cooling, and space – or getting another few years out of it.

"A major hardware refresh, like moving to a new form factor (e.g. from rackmount to blade chasses) and consolidation are typically driven by a bigger IT initiative, like 'Avoid having to build a new data center,'" says ESG's Bowker.

Swap server utilization for hardware lifetime: Some companies have gone to a full virtualized infrastructure, and ended up with servers at 80%, says Info-Tech's Sloan. Virtualization increases the burn rate of servers. “They won't last five years, maybe only three or two, because the hardware often isn't engineered to be driven at that more fully-utilized rate," he says.

Don't replace too frequently just for tech improvements: Replacing server hardware too frequently just to get newer CPUs isn't your best strategy, notes Insight 64's Brookwood. "Although CPU performance increases 10-30%, buying a new machine to replace last year's won't give you that big a boost, unless you also make other changes, e.g., from rotating storage to solid state,” he says. “You'll get a big bang for your buck changing or refreshing processor elements every two years – but also look at your network and storage."

"CPU vendors do an architecture refresh approximately every two years," says HP's Gromala. "The year in between will have see new features to provide a speed increase – higher GHz processor speeds, more cores, faster I/O, and the like."

Don’t put all your budget into maximizing CPU muscle: More powerful CPUs aren't necessarily the best refresh path, says Insight 64's Brookwood. "For some applications, people are shifting to low power processors instead of the higher powered ones. For example, instead of using Intel Xeon or AMD Opteron CPUs, you can fit 512 Atom CPUs in a 10u chassis, and get much better efficiency. This can be a match for workloads that don't massively scale-up."

Also, depending on your applications, don't look just at CPUs in terms of processing power, Brookwood says; also look at the Graphic Processing Unit (GPU). "With GPUs, floating-point-intensive workloads like resource recovery, scientific applications, or energy companies' ‘where to drill’ algorithms that need to process huge amounts of geographic sounding data can take up to months to run. Using the floating-point resources in GPUs can reduce that time to days or weeks."

Another performance-pumper can be investing in primary storage technology. "A solid state hybrid drive achieves can improve overall performance for a small investment than upgrading your processor, adding DRAM or other traditional performance-enhancing strategies," claims storage vendor Seagate.

Consider "Agility" and other business benefits: Trying to eke another year or two of service from a given machine may cost your business far more than slight hardware savings. Even if the TCO case isn't compelling, the business ROI for doing a server refresh may be.

"In many cases, we see companies shorten their server lifecycle because of the ROI to their business in terms of IT agility, being more productive, and being able to be more competitive," says HP's Gromala. "They are starting to use IT as a business agility tool, and for that, they may need the latest technology. A company may move from five years to three, or from three to two years." However, Gromala notes, "Not too many companies go to a lifecycle of less than 24 months."

Do a "cascade-down" refresh: In some cases, companies keep hardware for the original timeframe, but the lifecycle for its original purpose may be shortened.

Darin Stahl, Lead Analyst at Info-Tech Research Group, reports, "Large enterprise data centers are not refreshing faster than the CAPEX for existing servers is depreciated. In large enterprise datacenters, Info-Tech still sees a two to three year refresh cycle among its customers for the server hardware supporting 'Gold' service delivery with a cascade of hardware downwards to 'Silver' and eventually 'Bronze' tiers of infrastructure providing services delivery for lower business impact services. The cascading movement of servers into lower tiers means that some of the server hardware might take as long as four to five years before the server is finally exited out of the infrastructure."

There are, of course, other triggers to cause your company to say, “Okay, it’s time to break out the checkbook.”

Among the other factors that might accelerate server refresh in large enterprise environments, says Info-Tech's Stahl, are:

• Data Center relocation. Building a new data center or significant renovations to an existing facility may trigger early server refresh to successfully relocate service delivery from one facility to another while mitigating risk.
• Disaster Recovery. A revisit of the disaster recovery plan may trigger an early refresh of server hardware when establishing the mirrored environment.
• "Magic Moments." Stahl explains, "'Magic Moments' for IT occur when significant infrastructure changes are required to support business needs. Roll-up or upgrade of large enterprise application platforms, new lines of business, mergers and acquisitions all are potentially accelerants to the Infrastructure Strategy including frequency of refresh."

More importantly, review your IT infrastructure and requirements, and new product capabilities, on a regular basis.

"If you did 'sizing' of your applications two or three years ago for virtualization, do it again," advises Info-Tech's Sloan. "Some of the bottlenecks have gone away. For example, you can devote more RAM to a processor."

"For servers, I recommend that every year there be an optimization plan of record," urges Bruce Michelson, distinguished technologist at HP. "If your optimization plan for 2012 looks a lot like the one from 2011, you're probably missing a lot of opportunity, because not only the new technology that has becoming available, but also the service delivery approaches of cloud computing and virtualization.

"If you want to move a higher-tier workload to the environment, that requires access to more CPU, higher I/O throughput, more memory, and taking advantage of current features of the virtualization, you need to be aware of what the new server architecture has to offer," says ESG's Bowker.

"The new economics of IT provisioning has three major motivations," says HP's Gromala. "One, productivity gains from doing refresh to more powerful systems, and through better tools for to improve operations. Two, the sustainability in terms of power and cooling requirements versus resources. And three, being able to ride the innovation curve to improve business agility and competitiveness."