Mentioning Linux and server consolidation in one sentence brings up a host of different opinions, ranging from...
"What a great idea!" to "What, are you crazy?" But the fact is, server consolidation is a hot topic -- as is anything with the potential to save money these days. Add Linux, and people think that the opportunity to cut cost through space consolidation, workforce reduction and software savings could be huge.
The truth is, server sprawl is certainly a problem with many companies, says Jon Phelps, vice president and research director at Gartner Inc., in Stamford, Conn. "Today a typical data center may have a mainframe, but what fills the room is hundreds of Unix and Intel servers," he says.
But while the concept of server consolidation shows promise, there are a variety of issues -- from software licensing glitches to capacity management -- that stand in the way of immediate mainstream acceptance. Moreover, it is difficult to predict how just beneficial server consolidation will be to the IT budget. "Savings could range anywhere from millions of dollars to having server consolidation actually cost extra money," says Phelps. "There are too many factors to arrive at a hard average."
At present, consolidating servers to run on Linux comes in two different flavors:
Linux on the mainframe: The idea here is to partition an IBM zSeries into thousands of virtual Linux machines running Intel- or Unix-based applications under zVM. Talk about server consolidation! The downside is that you have to have a spare IBM mainframe handy to do so, as well as people with both Linux and OS/390 knowledge. It also means that all the Intel or Unix-based servers that used to run your enterprise applications are left to gather dust.
Linux on Intel Servers: The less costly alternative that's just beginning to surface is to consolidate your servers and carve them up using partitioning software such as those products available from SW Soft and VMWare. This helps put an end to server sprawl w/out tossing the baby out with the bath water, so to speak.
Although both of these schemes are attractive to those sick of running and maintaining thousands of underutilized servers, consolidation also brings some implementations issues to light.
Capacity management: Most Unix- or Windows-based servers are utilized at numbers ranging from 5-15%, says Jan-Rainer Lahmann, an eServer architect at IBM. "If you can consolidate onto systems capable of sharing processors, the capacity can be managed much more efficiently," he says. But Lahmann cautions companies not to cut too close to the bone: "You still need to have enough extra capacity to handle peaks." Other companies have different cycles, he says, although virtual servers can also share processing power according to workload. Lahmann also adds that there are several aspects to this: The peaks on each server are usually much higher than the average load (3x to 7x is very common). But when looking at all servers, not only at a single one, you find that the peaks occur at different points in time. So if you are able to share (or shift) processor resources between (virtual) servers, the amount of capacity that has to be provided to handle the peaks is by far not that high. In many cases, we found that 30% white space is a good number.
Unix servers typically peak at 65-80%. Taking the typical peak factor mentioned before, you then end up with 5-15% average utilization. Windows/Intel is even below Unix. On mainframes this is different because of the mixed workload that is being run on a single piece of hardware. This leads to higher utilization numbers -- 70% average and 85-95% sustained peak (over several hours) is common.
- Availability: Lahmann gives the example of Web servers, which generally scale horizontally. "If you consolidate onto one machine and that machine fails, the whole application or Website could go down. Before, with multiple machines, if one failed it wasn't a big problem. But if you reduce to one machine or a smaller number, planning for availability becomes very important."
- Software licensing: "ISVs in the Unix/Linux world are not used to running in a consolidated environment," says Phelps, so check your licensing agreement before consolidating. Why? Phelps gives the following example: "Say I was going to run eight different jobs under a single copy of an OS," he says. One of the applications runs on a two-way box now but I'm consolidating it onto a 32-way box. If I have user- or enterprise-based software licensing, it's not a problem. But if I have capacity based licensing, it's a problem because instead of running on a two-way I'm now running on a 32-way. I've had a couple people who couldn't consolidate because the software costs would wipe out consolidation savings."
Although the idea of consolidating servers under Linux -- whose open source code isn't controlled by any technology vendor -- certainly has allure, some analysts question whether Linux itself will actually be a money saver. After all, most corporations won't download the source code from the Internet. Instead, they'll buy Linux from distribution outfits like Red Hat or SuSe, for the added benefit of support services and care and feeding of the Linux code. So Linux in the data center will likely not be free. The real cost-cutter in this equation is the hardware, says Laura DiDio, an analyst at Yankee Group in Boston, Mass. "It's not Linux that's going to make server consolidation more cost effective, it's the hardware consolidation," she says.
Besides, DiDio contends that Linux does not yet have a great deal of traction in the data center. "At this point, I don't see anything but Unix playing a role in the data center for the next year or two," she says. "There is going to be a new wave of applications, such as Web Services, that will drive the need for more robust server consolidation, but I don't think we're going to be at that point for four or five years."
Either way, server consolidation could translate into job reductions, says Phelps. At companies that opt to shrink their servers onto fewer boxes, or consolidate them into one physical location, "There is a potential for some to lose jobs," he says. Phelps advises IS staffers to boost their value by learning skills that have allure in a consolidated environment. "Learn to understand partitioning and workload management," he says. "And start thinking about how to design applications that will run in an environment where other applications share the space."
Carol Hildebrand is a freelance writer based in Massachusetts.
>> Join Search390.com and Laura DiDio of Yankee Group for a webcast on server consolidation, Tues, December 17 at 2:00 PM EST. Laura will discuss the pros and cons of server consolidation, as well as offer her insights on various platforms.
- Search390.com's Best Web Links on Operating Systems contain a bunch of resources on server consolidation and various platforms.
- Check out this white paper on "The why and how of server consolidation."
- See Search390.com's past Featured Topic on server consolidation.