The jury is in. After years of experimentation with Linux in the enterprise, customers, analysts, and vendors are starting to sing a consistent tune about where Linux makes financial sense and where it doesn’t.
Although Linux is often thought of as a free alternative to established OSes such as Windows and proprietary Unix is it really cheaper when you add the costs of acquisition, migration, operation, and support? In other words, is the total cost of ownership (TCO) of Linux really lower than that of Unix or Windows?
The simple answer is this: The more fully an enterprise adopts Linux across its infrastructure, the more financial leverage it is likely to get out of upfront investments in the OS. Those investments, which can be considerable, include Linux training and tools, and the costs of migrating from a Unix or Windows environment. And that financial leverage is improving steadily as better management tools, more third-party vendor support, and more skilled Linux system administrators arrive on the market.
But the cost and benefit of switching are not equal for everyone. Unix-to-Linux migrations typically make financial sense because retraining costs are minimal while hardware acquisition costs drop significantly. In greenfield situations or for shops running a mish-mash of OSes, Linux can be a total financial no-brainer. But Windows-to-Linux migrations are more of a toss-up due to higher retraining and conversion costs, and lower hardware acquisition cost savings.
ROI or TCO?
Figuring out the TCO of Linux is not for the faint of heart. Several IT execs told us they skipped the exercise because the model would have been very complex, with too many unknowns and assumptions.
“It’s really hard to break Linux TCO down. It’s such a fluid environment,” said Ray Duncan, M.D., technology director at Cedars-Sinai Medical Center.
“It’s hard to get a side-by-side comparison.”
Laef Olson, an economist by training, decided to build a TCO model.
“I just sat down with a blank spreadsheet and started making assumptions,” said Olson, currently vice-president of technology for Cars.com, a division of Classified Ventures.
“Annual traffic growth. What do I pay for Web server CPUs? How many sys admins do I need per physical machine, and where will tools discount that down the road?”
Olson calculates that migrating to Linux would yield “a seven-figure savings” for Cars.com over a four-year period, assuming 20 per cent to 25 per cent annual traffic growth and a cost of capital (which he declined to reveal). But he said that his model contained a few big “ifs”, such as the availability of Linux OS support from key third-party vendors.
“A lot of it has to do with where you think the industry is going,” Olson said.
Another important question is whether to focus on TCO or return on investment (ROI) when evaluating Linux. Most industry studies have focused on TCO — the all-inclusive costs of running specific workloads such as Web serving, file and print, and security applications. But that is changing. The Robert Frances Group this year switched from doing Linux TCO analysis to ROI analysis.
“TCO doesn’t examine what you could save, what flexibility it will provide you, how much money you could make,” a senior analyst at the Robert Frances Group, Chad Robinson, said. “Those are the reasons you deploy something.”
Moreover, adopting Linux usually includes one-time, upfront migration costs such as retraining and application integration, costs that are included in TCO but can only really be evaluated by looking at ROI.
“If TCO triples but ROI doubles then it should be deployed,” Robinson said. “You end up saving in the long run.”
No matter which model you use, the financial benefit of switching to Linux from Unix or Windows is driven by four main cost categories: acquisition, migration, management, and support. Here are highlights of key issues in each of these areas.
Hardware and software costs
The benefit of replacing expensive RISC processor-based Unix hardware with commodity Intel boxes is one of the biggest factors driving Linux adoption.
“Discount retailing’s a tight business, and we’re wicked cheap,” Burlington Coat Factory CIO, Mike Prince, said.
Prince has deployed about 2400 Linux machines over the past three years, replacing many Unix systems.
“Instead of having a superhorse you have a team of horses — you don’t have to have this genetic (RISC) wonder,” he said. It’s an easy calculation.
“Moving Unix workloads to Linux is a no-brainer because of the Intel economics,” principal analyst at Forrester Research, Ted Schadler, said. “If you look at the all-in cost of deploying Unix on RISC versus that same workload on an HP or Dell box, it’s between a 5K and 25K price improvement.”
On the software side, the cost differential is less clear. Studies by IDC, Meta Group, and Robert Frances Group find that Linux license costs are lower than those of Windows. But some analysts argue it’s not an apples-to-apples comparison. “When you’re building apps,” Forrester’s Schadler said, “it’s not a Windows versus Linux decision. It’s a Java-on-Linux versus Windows decision. Microsoft bundles a lot of stuff into Windows, into SQL Server, into the .Net framework — if you’re looking to build a generic app and deploy it at an all-in price point, Windows is going to win hands down because you get so much bundled in.”
When contemplating a move from Unix or Windows to Linux, enterprises should take a hard look at the one-time migration costs. One of the biggest expenses is training systems administrators to get up to speed on Linux.
“You can correlate systems knowledge with age,” said Avery Lyford, CEO of Linuxcare, which develops management software for Linux environments. “It’s a gross generalisation, but if you talk to someone in their 20s, they know Linux; in their 30s, they know Microsoft; in their 40s, Unix; in their 50s, big systems like VMS [Virtual Memory System].” So in theory, Lyford said, you could gauge your Linux migration costs by figuring out the average age of your system administrators.
In reality, Unix skills are closer to Linux skills, which lowers the cost of migrating from Unix to Linux.
“If you’re a pure NT shop, and you don’t have any Linux skills, then the barrier is a heck of a lot higher,” Lyford said. “All the Unix people instantly get Linux.”
Those with Unix skills who don’t catch on right away can easily download and try it at home.
“You get to build the skills up for free on your own spare-time nickel,” Lyford said.
The ability to freely download Linux also makes it easier to prototype potential deployments, a gap that Microsoft is aiming to equalise.
“I need to make it easier for people to try to do things on my stuff, to try to build a scenario or an environment,” Microsoft’s general manager of platform strategy, Martin Taylor, said. “One of the issues that causes people to not take a full picture on (Linux costs) is they download something for free and they invest time to get it where they want it. They don’t fully account for the time and effort it took to even get their model scenario up and running.”
Other migration costs include code that may have to be rewritten, data that must be migrated, integration work to back-end systems, and software that must be purchased to replicate a capability that already exists on the platform Linux is displacing.
“There were a lot of costs I didn’t expect — hidden migration costs,” Duncan said.
During the migration from NT to Linux, his staff insisted that because they had been running RAID disk mirroring and striping on NT they should buy SCSI RAID controllers for the Linux servers.
“It was like $1000 per box extra that I hadn’t planned on,” he said.
One-time migration costs will be lower if your application already runs on top of middleware that will easily port to Linux, such as an application server.
“Modern development practices and cycles are helping reduce the migration cost,” Robinson notes.
By far the biggest cost in most Linux TCO studies is the staffing required for ongoing operational systems management. In a 2002 Windows vs. Linux study produced by IDC (and sponsored by Microsoft), staffing accounted for 62 percent of the total five-year cost for both environments. It also made up the biggest cost difference between the two, with Windows coming out ahead at a slightly lower cost.
Although most analysts believe Windows administrators cost less than Linux systems admins, the real debate centers around the cost benefits of consolidation, and the availability and quality of management software tools for the Linux platform.
“The operational savings vary wildly,” vice-president of Oracle’s Linux Program Office, Dave Dargo, said. “Customers [who] simply adopt Linux side by side with their other OSes probably won’t see savings.”
He makes the consolidation argument: If you move wholesale to Linux and buy or build robust management tools, you can save more with Linux than with proprietary Unix or Windows because Linux management tools and personnel skills can be leveraged across a wider target (servers, mainframes, and desktops). Because Linux developers have unfettered access to the Linux OS, fewer administrators are needed to manage more machines and greater workloads.
The only problem with this scenario for large Unix shops, director of enterprise hosting at EDS, DuWayne Mutchler said, was that some of those tools didn’t exist.
“The argument that Linux is cheaper to operate is one we struggle with,” Mutchler says. “The tools and processes and the automation capabilities have not evolved as far as they have for Unix, so we’re finding zero cost savings in moving to Linux.” Of EDS’s 50,000 servers, less than 2 pe rcent are running Linux.
Duncan has also found some key cost-saving tools missing, although he’s generally happy with his organisation’s migration to Linux.
One management cost area where Linux seems to consistently trump Windows is the cost of managing security. A big driver for Cedars-Sinai’s switch to Linux was “the tremendous amount of churn we have on our NT servers” due to hot fixes, service packs, and so on, Duncan said.
“We did an analysis of the amount of time we were spending tweaking NT servers, and it really was kind of terrifying,” he said. “ We should be able to set up a server and just leave it alone — we really got into Linux from that point of view.”
Linux was “virtually virus-free,” Burlington Coat Factory’s Prince agreed, and it was “pretty difficult for people to screw up their systems.”
The final major cost item in the debate over Linux TCO is support. Linux proponents claim Linux support is cheaper and available from a more diverse vendor population, and that Linux machines often run for years without so much as a reboot anyhow.
“Sun provides fantastic support on their equipment, and you pay dearly for it,” Cars.com’s Olson said. But when one of his new Linux machines breaks down, his staff either replaces it because that approach is cheaper than a fix, or they turn to what he calls “Google service” — looking on Web newsgroups and message boards for the solution.
Enterprises are overcoming their initial suspicion of this support method.
Users can choose support from their hardware or software vendor, distribution vendor, or a third party. “That’s possible with Windows, but [with] the Microsoft solution, providers can’t actually make a change to Windows for you,” Robinson said. “All of the [Linux] support vendors are equally empowered.”
Other Cost Contingencies
Although it is a factor hard to put into a spreadsheet, current and potential Linux users are also concerned about future cost risks associated with Linux’s ownership and development path.
Some worry that the Linux upgrade path will fragment as vendors develop proprietary flavours.
“The Linux industry needs to stay focused on what made it successful — the open source Linux platform,” CTO of Golden Gate University, Anthony Hill, said. If it fragments, “the economies of scale equation get diluted.”