When to upgrade

Insiders share rules of thumb about equipment life cycles.

Given the dearth of solid data on the subject, we surveyed the product-test experts in the Network World Lab Alliance and members of the Network World Advisory Board - executives in top IT spots in local Massachusetts organizations - for tips on when enough is enough.

The findings, as one respondent put it, confirm shop-honed intuition. The average life expectancy of hardware is three to five years, while equipment in fast-evolving markets, such as security, is being replaced in three years or less, and larger iron often hangs around for decades.

"We have some stuff that is eight or nine years old that just sits there and runs," says Richard Glasberg, director of enterprise communications for the commonwealth of Massachusetts. "But the rule of thumb for network infrastructure is three to five years. If you're trying to keep things humming along at 99.999% uptime, you'll question anything much older than that."

Besides upgrading equipment to stave off failure, another classic driver of change is the old software/hardware upgrade cycle, most famously witnessed in the PC/Windows realm.

"Advances in software mandates hardware changes, which serves as the platform until it no longer meets the needs of the software, so you change the hardware underneath and the dance continues," says Tom Henderson, principal researcher for ExtremeLabs in Indianapolis, a Lab Alliance partner.

Survey respondents expect Windows-based PCs to last 3.5 years on average, although some question whether that is about to change.

"There was a lot of thrash when we went from Windows 95 to 98 to 2000 and then to XP, with Microsoft upping the ante on how much resources you needed," says Lab Alliance member Joel Snyder, a senior partner at OpusOne, a consulting firm in Tucson, Ariz. "But things have been quiet since XP came out. XP-based systems may actually end up lasting five or more years - first, because there isn't a new operating system [in the immediate future], and second, anything that is XP-compatible probably has sufficient CPU and memory to run for quite a while."

Fast-evolving performance and capacity demands are, of course, at the root of many decisions to replace gear, from switches to servers and storage.

"It's not because your gear isn't up to spec or doesn't work anymore; it just doesn't do what you need it do anymore," Snyder says. "You might go buy a SonicWall firewall appliance for a T-1 at a remote office, and then two weeks later have your cable provider offer you 7M bit/sec."

Sometimes significant advances in a given tech sector provide the impetus, such as the arrival of 100M bit/sec Ethernet. While early adopters initially installed those big pipes to serve bandwidth-hungry segments, as LAN prices fell and application demand continued to advance, it didn't take long to reach the tipping point that led to wholesale network overhauls.

Technologies such as Gigabit Ethernet and 802.1X authentication may lead to a new round of infrastructure upgrades in the coming years.

VoIP is another new arrival that is spurring network overhauls. Data networks typically have to be spruced up before voice can be introduced into the traffic mix with any confidence. But while IP PBXs promise many advances, longevity - when compared with the machines they replace - isn't one of them. Our expert panel expects IP PBXs to last only 6.5 years on average, while traditional PBXs could be counted on for 8.5 or more years.

"Telecom products typically had long depreciation cycles because phones never changed, the software never changed and the application never changed," Henderson says. "But that's all changing with VoIP. I've seen some compelling new applications for Cisco and Avaya phones, but of course only if you have the phones with the cool 640x480 color LCD display. Phone features are evolving, which is cutting down on the useful lives of the equipment."

Sometimes the arrival of a new technology will encourage vendors to try to spur migration by driving maintenance costs for the old gear through the roof. "We had to kill our Lucent PBX even though it was only 3 years old because the upgrade cost was almost equal to the original capital expenditure," Henderson says.

Vendors typically say they will support a device for two years after they stop selling it, Glasberg says. So if you tack those two years on to a normal life expectancy of five years, you get an outside range of up to seven years.

In the later stages of that cycle, vendors offer trade-in allowances as an inducement to swap up, and that can lead to some tough decisions. Jump too early, and you don't get the most out of your original investment; jump too late and you minimize the allowances or miss them altogether, Glasberg says.

Luckily, the timing usually works out, he says. "When a vendor is ready to retire something, you're probably ready to get rid of it."

Security evolution

Security products are probably not the type of equipment that buyers will hold onto for seven or more years.

Security gear gets obsolete for two reasons, Snyder says. "One, the loads we put on them increase because bandwidth usage continues to go up, and two, we stretch them by loading up more applications."

As an example of the latter, Snyder says if you bought a firewall and later decided you wanted to use virus scanning, it might not be fast enough for the new load.

As vendors add more applications and functions to their gear, the performance tends to drop, encouraging people to consider new options. Security gear in general seems to be on a faster-than-normal upgrade cycle, Snyder says.

In some cases that cycle is simply driven by technology advances. A good example is an intrusion-detection system (IDS ). While IDS provided many new capabilities, intrusion prevention provided even more.

Big boxes

At the other end of the life-cycle spectrum you have big iron, such as large minicomputers and mainframes.

Even with new machines available for a fraction of the cost, customers are reluctant to replace this gear because the operating systems and applications are dyed-in-the-wool.

"There are still folks out there running VMS," Snyder says of the venerable Digital Equipment Corp. operating system. "It's not because they are in love with it, but because they have an app everyone is comfortable with and the performance hasn't degraded to the point where it's worth going through that horrible forklift upgrade."

However, with transaction volumes increasing and the size of transactions going up, companies may face the need to replace some big boxes sooner rather than later, Snyder says.

Will the upgrade conveyor belt ever stop? Probably not, but it may slow down as the technology becomes more commodity in nature, and fundamentals such as security are baked in from the start and we make progress toward true utility environments.

Until then, lace up your sneakers.